Participant global self‐assessment of acne improvement

A total of 39 trials addressed this outcome. Of these, 33 used Likert or Likert‐like scales, with categories/points ranging from 3 to 7. One trial used a 100‐mm visual analog scale (Miyachi 2016). In the other five studies, it was unclear which method was used (Chantalat 2006; Cunliffe 2002; Makino 2015; Tung 2014; Zeng 2012). This outcome was assessed at the end of treatment duration in all of these trials (ranging from 4 weeks to 16 weeks), with repeat measurements conducted in four trials (Hayashi 2018; Langner 2007; Pariser 2014; Schaller 2016). All four trials evaluated this outcome at weeks 2, 4, 8, and 12, and two trials additionally measured the outcome after the first week of treatment (Hayashi 2018; Langner 2007).

Withdrawal due to adverse events

A total of 68 trials reported how many participants withdrew due to adverse events. This outcome was monitored throughout the whole course of the trials.

Investigator‐assessed absolute or percentage change in lesion counts

A majority of trials (97/120) provided the number of absolute or percentage change in lesion counts for at least one lesion type. All of these trials evaluated the change from baseline at the end of treatment duration (ranging from 3 weeks to 52 weeks), among which 91 trials repeatedly measured the outcome, with the intervals between two measurements ranging from one to four weeks. It was unclear whether repeated measurements were conducted in the other six studies, in which only the last measurement was reported (NCT01044264; NCT01138514; NCT01231334; NCT02073461; NCT02465632; Study 152).

Percentage of participants rated 'clear' or 'almost clear' on the Investigator Global Assessment (IGA) scale of acne severity

A total of 25 trials assessed this outcome. The outcome was assessed at the end of treatment duration in all of the trials (ranging from 6 weeks to 24 weeks). Of 22 trials reporting repeat measurements of this outcome, 15 trials provided results for each measurement (Dreno 2011; Dréno 2018; Eichenfield 2013; Gold 2009; Gold 2010; Gollnick 2009; Hayashi 2018; Jawade 2016; Kaur 2015; Kawashima 2014; Kawashima 2015; Shafiq 2014; Thiboutot 2002; Thiboutot 2007; Zeichner 2013), and the other seven only presented the results for the last measurement (Bowman 2005; Fleischer 2010; Jackson 2010; Jones 2002; NCT00713609; Pariser 2014; Thiboutot 2008).

Change in quality of life

Only eight trials reported this outcome. Instruments used in these trials varied but included the following: Skindex‐29 (Dhawan 2013; Guerra‐Tapia 2012), Skindex‐16 (Hayashi 2018), Children's Dermatology Life Quality Index (Eichenfield 2013; Schaller 2016), acne‐specific quality of life (Chantalat 2006; Pariser 2014), Dermatology Life Quality Index (Schaller 2016), and Cardiff Acne Disability Index (Tabasum 2014). All of these trials assessed this outcome at baseline and at the end of treatment duration (at 6 or 12 weeks). Two of these trials additionally measured the outcome after two‐week treatments (Chantalat 2006; Guerra‐Tapia 2012), and one repeated measurements at weeks 1, 2, 4, 8, and 12 (Hayashi 2018).

Reduction in C acnes strains

This outcome was reported in eight trials, all of which assessed the change in total C acnes strains from baseline to the end of treatment (at week 12 or 16) (Cunliffe 2002; Eady 1996; Jackson 2010; Kawashima 2015; Langner 2007; Langner 2008; Leyden 2001b1; Leyden 2001b2). The change in clindamycin‐ and erythromycin‐resistant C acnes strains was also measured in six (Cunliffe 2002; Jackson 2010; Kawashima 2015; Langner 2007; Langner 2008; Leyden 2001b2) and four (Eady 1996; Jackson 2010; Langner 2007; Langner 2008) of these trials respectively. Five trials conducted repeat measurements with an interval varying between one and eight weeks (Cunliffe 2002; Eady 1996; Jackson 2010; Leyden 2001b1; Leyden 2001b2).

Percentage of participants experiencing any adverse event

A total of 96 trials provided the number of participants with any adverse events. This outcome was monitored throughout the whole course of the trials.

Main pooled analyses: BPO versus placebo or no treatment

Overall, the data for long‐term outcomes from three trials were complete to be pooled (Gold 2009; Jawade 2016; Leyden 2001a), and a total of 1111 participants in the BPO group and 1123 in the placebo group were included. The treatment duration was 10 weeks and 12 weeks in these trials. The outcome was assessed via a 6‐point or a 5‐point Likert‐like scale. The pooled risk ratio of treatment success was 1.27 (95% confidence interval (CI) 1.12 to 1.45; I² = 58%; Analysis 1.1), which shows some difference in this outcome between groups in favour of BPO treatment.

The risk ratio of treatment success for the medium term was 2.70 (95% CI 1.68 to 4.34; I² = 0%; Analysis 1.2) from two trials (Ozgen 2013; Papageorgiou 2000). Study authors used two different 6‐point scales to assess the outcome.

Subgroup analyses: co‐intervention

For Analysis 1.1 and Analysis 1.2, the test for subgroup differences was not statistically significant, with P values of 0.42 and 0.71, respectively. This suggests that effects of BPO did not differ by the co‐intervention.

Studies not included in the meta‐analysis

For the long‐term outcome, we did not include the other eight trials in the main analysis because the definitions of treatment success used were different from that in our review (Borglund 1991; Draelos 2002; Gollnick 2009; Thiboutot 2007; Thiboutot 2008; Tschen 2001; Tucker 1984; Zeichner 2013), but sensitivity analysis including these trials suggests similar results (risk ratio (RR) 1.32, 95% CI 1.20 to 1.45) to our main analysis in favour of BPO treatment (Appendix 9).

This outcome was assessed in the 10‐week trial comparing BPO plus meclocycline versus meclocycline monotherapy (Borglund 1991), but it is unclear how the outcome was assessed. It was reported that participants rating treatments as "good to excellent" accounted for 55% and 49% in the two groups, respectively, with an RR of 1.13 (95% CI 0.70 to 1.80).

This outcome was assessed via a 5‐point Likert‐like scale (from highly favourable to highly unfavourable) in one 12‐week parallel trial (Draelos 2002), with each group comprising 89 participants (BPO plus tazarotene versus tazarotene). At week 12, a significant difference was found in the proportion of participants rating the improvement as highly favourable or favourable (92% versus 73%; P < 0.05) in favour of BPO treatment.

In a 12‐week trial (Gollnick 2009), 415 participants in the BPO group, 418 in the placebo group, 419 in the BPO/adapalene group, and 418 in the adapalene group were assessed for this outcome via a 6‐point Likert‐like scale. The risk ratio for self‐reported complete or marked improvement was 1.16 (95% CI 1.01 to 1.34) in favour of BPO treatment.

A similar 6‐point Likert‐like scale was used in another 12‐week trial in which 149 participants were treated with BPO, 71 with placebo, 149 with BPO/adapalene, and 148 with adapalene (Thiboutot 2007). The risk ratio for self‐reported complete or marked improvement was 1.53 (95% CI 0.87 to 2.69) with no significant difference.

Thiboutot 2008, with 809 participants in the BPO group, 395 in the placebo group, 797 in the BPO/clindamycin group, and 812 in the clindamycin group, assessed the outcome after 12 weeks using a 7‐point Likert‐like scale. The risk ratio for self‐reported clear or almost clear was 1.58 (95% CI 1.35 to 1.86) in favour of BPO treatment.

In a 10‐week trial, a 7‐point Likert‐like scale (from 3 = much better to ‐3 = much worse) was used to assess the outcome (95 participants treated with BPO, 48 with placebo, 95 with BPO/clindamycin, and 49 with clindamycin) (Tschen 2001). The risk ratio for being "much better" on this scale was 2.58 (95% CI 1.42 to 4.68) in favour of BPO treatment.

In one 10‐week parallel trial with 24 participants on BPO plus clindamycin treatment and 29 on clindamycin alone, 96% and 95% of participants felt their acne had improved at week 10, respectively, with no statistical difference (Tucker 1984). However, it is unclear how this outcome was assessed.

Zeichner 2013 included 20 participants in each group (BPO plus tretinoin/clindamycin versus placebo plus tretinoin/clindamycin) and assessed this outcome using a 6‐point Likert‐like scale (0 = clear to 5 = severe). At week 12, no significant difference was found in the proportion of participants who scored 0 or 1 (5/20 versus 2/20; P = 0.24).

The long‐term outcome was also assessed in a 12‐week parallel trial comparing BPO plus tretinoin versus placebo plus tretinoin, with completion of follow‐up in 56 of 87 participants (Shalita 2003). No information about the criteria for the measure was available. For this outcome, study authors reported that participants in both groups rated their improvement as mild.

Vasarinsh 1969, with varied treatment duration, assessed the outcome using a 4‐point Likert‐like scale (from worse = ‐1 to greatly improved = 2). The average score was 0.66 and 0.53 for the BPO and placebo groups, respectively. This trial also compared BPO/sulphur with sulphur, and the average score was 1.15 and 0.75, respectively.

Main pooled analyses: BPO versus placebo or no treatment

For the long‐term period, risk of withdrawal due to adverse events was higher in the BPO group (RR 2.13, 95% CI 1.55 to 2.93; 13,744 participants; I² = 0%; Analysis 1.3). According to the funnel plot (Figure 4) and Egger's test (P = 0.753), no evidence suggests potential publication bias. Erythema, pruritus, and skin burning were the most common causes of withdrawal.

Figure 4

---

---

Funnel plot of comparison: 1 BPO versus placebo or no treatment, outcome: 1.3 Withdrawal due to adverse effects (long‐term data).

In terms of the medium‐term outcome, one trial found no withdrawals due to adverse effects (Mills 1986), another trial observed that only one participant discontinued BPO treatment (due to dermatitis) (Burke 1983), and one trial reported that only one participant on BPO plus nadifloxacin withdrew because of a severe burning sensation (Ozgen 2013). These three trials suggested no clear differences between the two groups (RR 2.92, 95% CI 0.31 to 27.44; 202 participants; 3 studies; I² = 0%; Analysis 1.4).

Subgroup analyses: co‐intervention

For long‐term data, the test for subgroup differences showed a borderline statistically significant difference when different co‐interventions were used (P = 0.05; Analysis 1.3). This potential difference came from the comparison between BPO plus tazarotene and tazarotene. In this comparison, BPO treatment was associated with non‐significantly decreased risk of withdrawal from the trial, but in other comparisons the association was opposite.

For medium‐term data, we did not observe significant differences between subgroups (P = 1.00; Analysis 1.4).

Studies not included in the meta‐analysis

Borglund 1991, the long‐term trial, reported that no participants discontinued BPO plus meclocycline or meclocycline monotherapy due to adverse events (Appendix 10).

Two studies, which assessed medium‐ and short‐term outcomes, respectively, were not included in the meta‐analysis. Papageorgiou 2000 reported that two participants discontinued treatments because of flare‐up. However, it is unclear what treatments the participants received previously.

The other trial was the only one that focused on the short‐term outcome. Of 61 participants enrolled (30 in the BPO plus tretinoin/clindamycin group and 31 in the placebo plus tretinoin/clindamycin group), no participants discontinued treatment because of adverse events during the four‐week period (Draelos 2010).

Main pooled analyses: BPO versus placebo or no treatment

For long‐term treatment duration, the absolute change in TLs pooled from the three trials was ‐10.73 (95% CI ‐15.68 to ‐5.78; 3230 participants; I² = 84%; Analysis 1.5), in favour of both BPO monotherapy and add‐on treatment (Eichenfield 2011; Kawashima 2014; Xu 2016). The three trials also suggested significant absolute change in ILs and NILs among participants on BPO treatment, with pooled mean differences of ‐3.50 (95% CI ‐6.33 to ‐0.67; 2635 participants; I² = 92%; Analysis 1.6) and ‐6.53 (95% CI ‐9.74 to ‐3.32; 2635 participants; I² = 75%; Analysis 1.7), respectively. The two studies were conducted in western and Asian populations, respectively, which may lead to heterogeneity in the reduction in acne lesions. The difference in percentage change in TLs was 10.29% (95% CI 3.39% to 17.19%; 1567 participants; 4 studies; I² = 72%; Analysis 1.8), which was pooled from four trials (Cunliffe 2002; Fleischer 2010; Miyachi 2016; Xu 2016). The counterpart in ILs and NILs from four trials was 17.22 (95% CI 4.98 to 29.45; 1588 participants; I² = 87%; Analysis 1.9) and 19.31 (95% CI 7.16 to 31.47; 1588 participants; I² = 80%; Analysis 1.10), respectively (Cunliffe 2002; Fleischer 2010; Lookingbill 1997; Xu 2016).

As for the medium‐term outcome, we found significant differences in percentage reduction in NILs (mean difference (MD) 18.42, 95% CI 1.39 to 35.45; 1212 participants; 3 studies; I² = 89%; Analysis 1.13) in favour of BPO treatment but not in TLs (MD 16.02, 95% CI ‐0.95 to 32.99; 252 participants; 2 studies; I² = 65%; Analysis 1.11) nor ILs (MD 13.83, 95% CI ‐0.22 to 27.88; 1212 participants; 3 studies; I² = 86%; Analysis 1.12).

These trials also reported the short‐term outcome. We found significant differences in percentage reduction in TLs (MD 10.00, 95% CI 2.26 to 17.74; 201 participants; 1 study; Analysis 1.14) and NILs (MD 24.65, 95% CI 4.23 to 45.07; 229 participants; 2 studies; Analysis 1.16) in favour of BPO treatment but not in ILs (MD 14.02, 95% CI ‐0.60 to 28.63; 1212 participants; 3 studies; I² = 91%; Analysis 1.15).

Subgroup analyses: co‐intervention

Most subgroup analyses suggested significant differences were common for BPO monotherapy and BPO combination (Analysis 1.6; Analysis 1.7; Analysis 1.8; Analysis 1.9; Analysis 1.10; Analysis 1.12; Analysis 1.13; Analysis 1.15). Reduction in acne lesions tended to be greater with BPO monotherapy than with BPO combination therapies. On the other hand, we did not observe the subgroup difference in absolute or percentage change in TLs regardless of treatment duration (Analysis 1.5; Analysis 1.11; Analysis 1.14).

Studies not included in the meta‐analysis

In the other trials, data on the absolute or percentage change in acne lesions were insufficient to summarise the effect size.

Three trials investigated BPO monotherapy. Leyden 2001a claimed that neither TLs (‐13.8 versus ‐1.3) nor ILs (‐5.9 versus 0.7) were significantly reduced in the BPO group when compared to the placebo group at week 10. However, another two trials did not show the difference. In Study 152, the percentage reduction in TLs, ILs, and NILs was 25.5% versus 20.6% (P value not reported), 33.5% versus 28.6% (P = 0.54), and 18.8% versus 15.4% (P = 0.49) for the BPO group (n = 70) and the placebo group (n = 37). Vasarinsh 1969 assessed the change in lesion counts for superficial (comedones and pustules) and deep (papules and cysts) lesions, respectively, using a 11‐point scoring system (decrease in lesions by 100% = 5, 76% to 99% = 4, 51% to 75% = 3, 26% to 50% = 2, < 25% = 1, no change = 0; increase by < 25% = ‐1, 26% to 50% = ‐2, 51% to 75% = ‐3, 76% to 100% = ‐4, over 100% = ‐5). The average score for superficial lesions was 0.55 and 0 for the BPO and placebo groups, respectively, and the counterpart for deep lesions was 0.69 and 0.53.

Four trials comparing BPO/adapalene with adapalene reported the percentage reduction in lesion counts (Gold 2009; Gollnick 2009; Miyachi 2016; Thiboutot 2007). Gold 2009 found a significant difference in the percentage reduction in TLs (60.7% versus 49.8%), ILs (66.0% versus 52.1%), and NILs (58.9% versus 52.9%) between groups (all P values < 0.05). In Gollnick 2009, the percentage reduction in TLs, ILs, and NILs was significantly higher in the BPO/adapalene group (68.8% versus 55.7%, 72.4% versus 61.9%, and 66.4% versus 54.3%, respectively) (all P values < 0.05). In Miyachi 2016, which compared BPO/adapalene gel with adapalene in participants (212 versus 107 participants), the percentage reduction in TLs was 82.7% versus 68.6%, with a significant difference (MD 14.6%, 95% CI 8.3% to 20.8%). The average percentage reduction in ILs and NILs was 80.9% versus 66.2% (P < 0.001) and 74.6% versus 61.1% (P < 0.001), respectively. Another trial found 51.0% and 35.4% reduction in TLs, 62.9% and 45.7% in ILs, and 51.2% and 33.3% in NILs for BPO/adapalene and adapalene, respectively (all P values < 0.001) (Thiboutot 2007).

Six trials compared BPO/clindamycin with clindamycin (Leyden 2001a; Study 152; Study 156; Thiboutot 2008; Tschen 2001; Tucker 1984). For the absolute change in lesion counts, in another trial, investigators claimed that both TLs (‐18.4 versus ‐6.2; P < 0.001) and ILs (‐10.1 versus ‐3.2; P < 0.001) were significantly reduced in the BPO/clindamycin group when compared to the clindamycin group at week 10 (Leyden 2001a). Tschen 2001 found ‐16.6 versus ‐14.7 in ILs for BPO/clindamycin (95 participants) and clindamycin (49 participants) (P = 0.034). In Tucker 1984, trial authors reported the absolute change in comedone, papule, pustule, and cyst lesions at week 10 separately: corresponding results were ‐21.0 versus ‐7.3, ‐11.3 versus ‐9.7, ‐8.3 versus ‐3.4, and ‐1.4 versus ‐2.3. For the percentage change in lesion counts, in Study 152, the percentage reduction in TLs, ILs, and NILs was 32.5% versus 23.5% (P = 0.002), 43.4% versus 39.8% (0.538), the 25.7% versus 11.2% (P < 0.001) for the BPO/clindamycin group (n = 73 participants) and the clindamycin group (n = 70). In Study 156, the percentage reduction in TLs, ILs, and NILs was 49.8% versus 33.3% (P < 0.001), 57.3% versus 48.6% (0.03), and 39.0% versus 18.0% (P < 0.001) for the BPO/clindamycin group (n = 96) and the clindamycin group (n = 96). For Thiboutot 2008, 797 participants were randomised to the BPO/clindamycin group and 812 participants to the clindamycin group. The percentage reduction was 47.9% versus 40.4% in TLs, 54.6% versus 46.2% in ILs, and 43.2% versus 36.2% in NILs at week 12 (all P values < 0.001).

Other comparisons for BPO as an add‐on treatment were limited. One 10‐week trial focusing on BPO/erythromycin versus erythromycin assessed the mean difference in the percentage reduction in lesion counts between two groups but reported only that the percentage reduction in comedone, pustule, papule, and inflammatory lesions seemed greater in the BPO/erythromycin group, with no significant difference (Chalker 1983). Two trials assessed the reduction in acne lesions for BPO plus adapalene versus adapalene (Cassano 2002; Fang 2002). Only an abstract was available for Cassano 2002, in which a total of 162 participants were randomised to one of the four groups: adapalene gel; adapalene gel; adapalene gel plus BPO gel 5%; and adapalene plus erythromycin gel 4%. With no information on the absolute or percentage reduction in lesion counts, trial authors claimed that all these treatments similarly led to a significant reduction in the numbers of papules, pustules, and comedones. The results of Fang 2002 were also presented in an abstract. A total of 150 participants were randomised to one of the two groups. At week 12, there was no significant difference in the percentage reduction in TLs (81.3% versus 68.9%), despite higher reduction found in the BPO plus adapalene group. One 12‐week parallel trial comparing BPO plus tazarotene with tazarotene reported that the percentage reduction in both ILs (39.7% versus 43.7%, P > 0.05) and NILs (42.7% versus 41.6%, P > 0.05) was similar between two groups (Draelos 2002). For the medium‐ and short‐term periods, both treatments reduced NILs by about 40% and 30%, respectively, with no significant difference. Their effects on reducing ILs were also comparable (31% versus 31% for the medium‐term and 24% versus 17% for the short‐term period). In a 12‐week parallel trial (Shalita 2003), researchers investigating BPO plus tretinoin versus placebo plus tretinoin observed that the mean number of IL counts was reduced from 9.4 at baseline to 3.9 at week 12 among participants on BPO, and from 9.3 to 6.6 among participants on placebo (P < 0.01). For NILs, there were no significant differences between treatment groups, although both groups had a significant reduction from baseline. One 10‐week parallel trial with 24 participants on BPO plus clindamycin and 29 participants on clindamycin alone reported the absolute reduction in comedone, pustule, papule, and cyst lesions from baseline to week 10 (‐20.4 versus ‐7.3, ‐5.4 versus ‐3.4, ‐11.4 versus ‐9.7, and ‐1.4 versus ‐2.4, respectively) (Tucker 1984). The significance of test results was not reported. As reported in one 10‐week parallel trial for BPO plus meclocycline sulfosalicylate versus meclocycline sulfosalicylate (Borglund 1991), the difference in the absolute reduction in papules, pustules, and comedones was ‐2, ‐1.6, and ‐2.6, respectively, for long‐term data. For medium‐term outcomes, the counterpart was ‐2.9, ‐2.4, and ‐1, respectively. For short‐term outcomes, the counterpart was ‐4.8, ‐2.1, and 1.3, respectively. All results were not statistically significant (all P values > 0.05).

In an 8‐week parallel trial with 47 participants on BPO plus nadifloxacin and 46 participants on placebo plus nadifloxacin (Ozgen 2013), the mean percentage reduction in ILs and NILs was 53.5% versus 22.08% and 34.8% versus 7.6% (both P < 0.001) at week 8. Vasarinsh 1969 compared BPO/sulphur with sulphur. The change in lesion counts in this trial was assessed for superficial (comedones and pustules) and deep (papules and cysts) lesions, respectively, using a 11‐point scoring system (decrease in lesions by 100% = 5, 76% to 99%=4, 51% to 75%=3, 26% to 50%=2, < 25% = 1, and no change = 0; increase by < 25% = ‐1, 26% to 50% = ‐2, 51% to 75% = ‐3, 76% to 100% = ‐4, over 100% = ‐5). The average score for superficial lesions was 0.81 and ‐0.70 for the BPO/sulphur and sulphur groups, respectively, and the counterpart for deep lesions was 0.91 and 0.30.

We also found some trials that comparing only lesions between groups at end of treatment, rather than the absolute or percentage change in lesions.

For BPO monotherapy, three trials reported long‐term data ‐ Hughes 1992 ‐ or medium‐term data (Burke 1983; Mills 1986); all supported the benefits of BPO treatment. The mean difference was ‐27.50 (95% CI ‐46.77 to ‐8.23) in ILs, and ‐28.20 (95% CI ‐45.04 to ‐11.36) in NILs (Hughes 1992). The mean difference was ‐16.00 (95% CI ‐18.25 to ‐13.75) in TLs, and ‐8.40 (95% CI ‐9.88 to ‐6.92) in NILs, as reported in Burke 1983. The pooled mean difference in ILs was ‐4.40 (95% CI ‐5.47 to ‐3.33; I² = 0%).

For BPO add‐on treatment, two trials presented long‐term data for BPO plus adapalene versus adapalene (Fu 2003; Korkut 2005). No significant difference was found for TLs (MD 0.11, 95% CI ‐4.94 to 5.17; I² = 0%), ILs (MD ‐0.52, 95% CI ‐2.15 to 1.10; I² = 0%), and NILs (MD 0.87, 95% CI ‐2.15 to 3.88; I² = 0%).

Main pooled analyses: BPO versus placebo or no treatment

The long‐term outcome was reported in 10 trials (Eichenfield 2011; Fleischer 2010; Gold 2009; Gollnick 2009; Kawashima 2014; Kawashima 2015; Thiboutot 2007; Thiboutot 2008; Xu 2016; Zeichner 2013), with an RR of 1.55 (95% CI 1.40 to 1.70; 10,399 participants; I² = 43%; Analysis 1.17). Counterparts for medium‐ and short‐term outcomes were reported in five studies (Gold 2009; Gollnick 2009; Kawashima 2014; Kawashima 2015; Thiboutot 2007), with an RR of 1.96 (95% CI 1.58 to 2.44; 5014 participants; I² = 35%; Analysis 1.18) and 2.43 (95% CI 1.78 to 3.32; 5014 participants; I² = 19%; Analysis 1.19), respectively.

Subgroup analyses: co‐intervention

In subgroup analysis, we did not observe that the effects of BPO can differ by co‐intervention, regardless of treatment duration (Analysis 1.17; Analysis 1.18; Analysis 1.19).

Studies not included in the meta‐analysis

This outcome was reported in only two trials (Eady 1996; Kawashima 2015). In Kawashima 2015, 34 (6.8%) in the BPO/clindamycin group and 59 (19 7%) in the clindamycin group had a sample of C acnes. Study authors reported that a small increase in the number of clindamycin‐resistant C acnes isolates was found across groups. Eady 1996 randomised 20 participants into the combination treatment group and 17 into the erythromycin group. BPO/erythromycin treatment was more effective in reducing total propionibacterial numbers than was erythromycin alone at week 12. Meanwhile, erythromycin‐resistant propionibacteria were found in seven participants treated with combination formulation and in 10 participants treated with erythromycin alone.

Main pooled analyses: BPO versus placebo or no treatment

Overall, we observed that BPO treatment can potentially result in higher risk of adverse events in the long‐term treatment period, with an RR of 1.40 (95% CI 1.15 to 1.70; 11,028 participants; I² = 72%; Analysis 1.20). According to the funnel plot (Figure 5) and Egger's test (P = 0.881), no evidence suggests potential publication bias. Most adverse events were mild to moderate. The most common adverse events across trials included skin dryness, facial pain, pruritus, dermatitis, erythema, and irritation.

Figure 5

---

---

Funnel plot of comparison: 1 BPO versus placebo or no treatment, outcome: 1.20 Percentage of participants with any adverse events (long‐term data).

Evidence for medium‐term or short‐term treatment duration was limited, with an RR of 1.67 (95% CI 1.08 to 2.59; 93 participants; 1 study; Analysis 1.21) and 0.13 (95% CI 0.02 to 0.94; 60 participants; 1 study; Analysis 1.22), respectively.

In the trial with varying treatment durations, excessive erythema and dryness occurred in five participants on BPO and two on placebo. One participant developed severe allergic sensitisation after BPO treatment (Vasarinsh 1969).

Subgroup analyses: co‐intervention

We did not observe that the occurrence of adverse events differed by co‐intervention (P for subgroup differences = 0.32).

Studies not included in the meta‐analysis

Two trials did not present the number of participants experiencing at least one adverse event but reported that the number was similar across study treatments (Borglund 1991; Gold 2009). Although adverse events were assessed in four trials exploring effects of BPO plus adapalene versus adapalene (Cassano 2002; Fang 2002; Fu 2003; Korkut 2005), three of them did not present the outcomes of our review interest (Cassano 2002; Fang 2002; Korkut 2005). We cannot calculate RR as the effect size for the comparison between BPO plus tretinoin and placebo plus tretinoin as investigated in one 12‐week parallel trial because the number of participants with at least one adverse event was not provided (Shalita 2003). Vasarinsh 1969 also assessed adverse events for the comparison between BPO/sulphur and sulphur, but the total number of participants with any adverse event was not reported. As mentioned in Kabir 2018, adverse events were noted during the trial but were not reported.

Main pooled analyses: BPO versus adapalene

Complete data for the long‐term outcome were available in five trials (Babaeinejad 2013; do Nascimento 2003; Gold 2009; Hayashi 2018; Jawade 2016), in which the outcome was assessed with a four‐category, a 5‐point, or a 6‐point Likert‐like scale. No difference in the probability of treatment success was found (RR 0.99, 95% CI 0.90 to 1.10; 1472 participants; 5 studies; I² = 55%; Analysis 2.1).

Subgroup analyses: co‐intervention

We found that effects of BPO versus adapalene may be altered by co‐intervention with clindamycin, with a P value of 0.03 (Analysis 2.1).

Studies not included in the meta‐analysis

We did not pool the results from the other three trials because data were insufficient to calculate the percentage of treatment success defined in our review (Appendix 9) (Gollnick 2009; Miyachi 2016; Thiboutot 2007). In Gollnick 2009, 415 participants in the BPO group and 418 in the adapalene group were assessed for this outcome via a 6‐point Likert‐like scale. The risk ratio for self‐reported complete or marked improvement was 0.89 (95% CI 0.59 to 1.34). A similar 6‐point Likert‐like scale was used in another trial (Thiboutot 2007), in which 149 participants were treated with BPO and 148 with placebo. The risk ratio for self‐reported complete or marked improvement was 0.88 (95% CI 0.63 to 1.22). Sensitivity analysis including both studies revealed similar results (RR 0.99, 95% CI 0.90 to 1.08). These results consistently suggested no difference in self‐assessed improvement between the two groups. In Miyachi 2016, where self‐reported improvement was assessed via a 100‐mm visual analog scale (0 mm = not completely satisfactory, 100 mm = very satisfactory), the average (± standard deviation) score at 12 weeks was 79.6 ± 22.0 mm for BPO and 74.4 ± 19.7 mm for adapalene.

Main pooled analyses: BPO versus adapalene

We did not observe significant differences in the long‐term withdrawal rate between BPO and adapalene, with an RR of 1.85 (95% CI 0.94 to 3.64; 3295 participants; 11 studies; I² = 0%; Analysis 2.2). As shown in the funnel plot (Figure 6) and on Egger's test (P = 0.407), no evidence suggests potential publication bias. For medium‐term data, no participants discontinued treatment in Stinco 2007.

Figure 6

---

---

Funnel plot of comparison: 2 BPO versus adapalene, outcome: 2.2 Withdrawal due to adverse effects (long‐term data).

Subgroup analyses: co‐intervention

No significant between‐group differences by co‐interventions were observed in subgroup analyses for the long‐term outcome (P = 0.60).

Main pooled analyses: BPO versus adapalene

In the trial comparing BPO/clindamycin with adapalene plus clindamycin among 349 participants (Hayashi 2018), the mean difference in absolute change in TLs, ILs, and NILs assessed after a long‐term treatment duration was ‐1.70 (95% CI ‐5.46 to 2.06; Analysis 2.3), ‐1.10 (95% CI ‐2.42 to 0.22; Analysis 2.4), and ‐0.60 (95% CI ‐3.65 to 2.45; Analysis 2.5), respectively, with no significant differences between the two groups. Counterparts reported in this trial were ‐0.70 (95% CI ‐5.09 to 3.69; Analysis 2.6), ‐1.90 (95% CI ‐3.44 to ‐0.36; Analysis 2.7), and 1.20 (95% CI ‐2.46 to 4.86; Analysis 2.8) for the medium‐term period and ‐4.70 (95% CI ‐9.39 to ‐0.01; Analysis 2.9), ‐2.60 (95% CI ‐4.29 to ‐0.91; Analysis 2.10), and ‐2.00 (95% CI ‐6.02 to 2.02; Analysis 2.11) for the short‐term period, respectively.

For the percentage change in TLs, ILs, and NILs, the mean difference was ‐2.63 (95% CI ‐18.42 to 13.15; 869 participants; 4 studies; I² = 98%; Analysis 2.12), ‐5.70 (95% CI ‐21.14 to 9.74; 806 participants; 4 studies; I² = 98%; Analysis 2.13), and ‐7.09 (95% CI ‐21.39 to 7.21; 806 participants; 4 studies; I² = 97%; Analysis 2.14). Counterparts for the medium‐term period reported in this trial were 0.56 (95% CI ‐5.04 to 6.16; 547 participants; 2 studies; I² = 33%; Analysis 2.15), 5.55 (95% CI 1.58 to 9.52; 689 participants; 3 studies; I² = 0%; Analysis 2.16), and 0.89 (95% CI ‐8.35 to 10.12; 689 participants; 3 studies; I² = 71%; Analysis 2.17), and for the short‐term period, 4.50 (95% CI 0.22 to 8.78; 547 participants; 2 studies; I² = 10%; Analysis 2.18), 9.12 (95% CI 4.98 to 13.26; 689 participants; 3 studies; I² = 3%; Analysis 2.19), and 6.18 (95% CI ‐1.80 to 14.15; 689 participants; 3 studies; I² = 64%; Analysis 2.20), respectively.

Subgroup analyses: co‐intervention

The percentage change in TLs, ILs, and NILs may differ by co‐interventions (Analysis 2.12; Analysis 2.13; Analysis 2.14; Analysis 2.17; Analysis 2.20). BPO monotherapy seemed to achieve greater reduction in TLs for the long term (Analysis 2.12) and in NILs for the medium (Analysis 2.17) and short term (Analysis 2.20) than adapalene monotherapy. On the other hand, greater reduction in ILs and NILs was observed with BPO/clindamycin versus adapalene plus clindamycin for the long term (Analysis 2.13; Analysis 2.14). We did not observe significant differences in the changes in any other acne lesions (Analysis 2.15; Analysis 2.16; Analysis 2.18; Analysis 2.19).

Studies not included in the meta‐analysis

We also found some studies that assessed the change in acne lesions but did not provide sufficient data for meta‐analysis. For the comparison between BPO monotherapy versus adapalene monotherapy, we can compare the lesions between groups at end of treatment from only two trials (Babaeinejad 2013; Korkut 2005): the mean difference was 3.62 (95% CI 2.02 to 5.22; I² = 0%) in TLs, ‐1.25 (95% CI ‐4.23 to 1.73; I² = 80%) in ILs, and 2.45 (95% CI 0.87 to 4.04; I² = 0%) in NILs. In Iftikhar 2009, no significant difference in the absolute reduction in TLs was found. The percentage reduction in TLs (35.6% versus 35.4%), ILs (43.6% versus 45.7%), and NILs (36.4% versus 33.3%) was similar between trial groups (Thiboutot 2007). In Miyachi 2016, which compared BPO with adapalene in participants (104 versus 101 participants), the percentage reduction in TLs was 73.5% versus 62.7%, with a significant difference (MD 10.80%, 95% CI 3.38 to 18.22%). The average percentage reduction in ILs and NILs was 78.9% versus 66.2% and 78.8% versus 61.1% (with no significance test results available), respectively.

Main pooled analyses: BPO versus adapalene

Overall, there were no significant differences in this outcome after a long‐term treatment duration (RR 1.12, 95% CI 0.95 to 1.32; 2512 participants; 5 studies; I² = 20%; Analysis 2.21), regardless of BPO used as monotherapy or add‐on treatment. A shorter treatment period seemed to find the higher probability of achieving 'clear' or 'almost clear' in the BPO group, as shown by the risk ratio for the medium term (RR 1.36, 95% CI 1.07 to 1.74; 2314 participants; 4 studies; I² = 0%; Analysis 2.22) and for the short term (RR 2.14, 95% CI 1.41 to 3.27; 2314 participants; 4 studies; I² = 0%; Analysis 2.23).

Subgroup analyses: co‐intervention

No evidence suggests that co‐intervention may alter the effects of BPO on this outcome for a long, medium, or short period (all P values for testing subgroup differences > 0.05; Analysis 2.21; Analysis 2.22; Analysis 2.23).

Studies not included in the meta‐analysis

Only one trial assessed the outcome using Skindex‐16 for the comparison between BPO/clindamycin versus adapalene plus clindamycin among 349 participants (Hayashi 2018). Trial results suggest no significant differences between the two groups for the long‐term period, with an MD of ‐0.17 (95% CI ‐0.37 to 0.03; Analysis 2.24). For medium‐ and short‐term periods, more benefits were observed in the BPO add‐on treatment group, with an MD of ‐0.22 (95% CI ‐0.42 to ‐0.02; Analysis 2.25) and ‐0.22 (95% CI ‐0.41 to ‐0.03; Analysis 2.26), respectively.

Main pooled analyses: BPO versus adapalene

Of the ten trials, seven provided the number of participants with adverse events, and pooled results from these trials show no significant difference in the long‐term outcome between the two groups (RR 0.71, 95% CI 0.50 to 1.00; 2120 participants; I² = 70%; Analysis 2.27). One trial, conducted exclusively in Japanese people, increased the heterogeneity in this meta‐analysis (Miyachi 2016). Most adverse events were mild to moderate and included skin dryness, peeling, burning, redness, nasopharyngitis, erythema, and application site pain.

The risk of adverse events for the short‐term treatment duration was substantially uncertain, with an RR of 3.00 (95% CI 0.13 to 68.26; 70 participants; 2 studies; I² = 0%; Analysis 2.28). No adverse events for the medium‐term period were reported in the trial, which included 40 participants in total (Stinco 2007). After the four‐week treatment, one of 15 participants treated with BPO plus clindamycin had hyperpigmentation, and none of 15 participants treated with adapalene plus clindamycin reported an adverse event (Dudhia 2015).

Subgroup analyses: co‐intervention

We did not observe that the long‐term outcome differed by co‐intervention (P value for subgroup differences = 0.44; Analysis 2.27).

Studies not included in the meta‐analysis

Another two trials mentioned only that no statistical differences in adverse events were found between the two groups (Gold 2009; Korkut 2005).

Main pooled analyses: BPO versus clindamycin

With 120 participants treated with BPO and 120 with clindamycin, the risk ratio of treatment success assessed via a 5‐point Likert‐like scale was 0.95 (95% CI 0.68 to 1.34; Analysis 3.1) (Leyden 2001a).

Studies not included in the meta‐analysis

We did not pool the results with those from the other three trials (Draelos 2002; Thiboutot 2008; Tschen 2001), given that the number of treatment successes defined in our review was not available (Appendix 9). However, sensitivity analysis including these studies showed a similar result (RR 1.05, 95% CI 0.97 to 1.15) to the main analysis.

Thiboutot 2008 randomly assigned 809 participants to the BPO group and 812 to the clindamycin group and assessed the outcome after 12 weeks using a 7‐point Likert‐like scale. The risk ratio for self‐reported clear or almost clear was 1.00 (95% CI 0.86 to 1.16).

In another 10‐week trial (Tschen 2001), a 7‐point Likert‐like scale (from 3 = much better to ‐3 = much worse) was used to assess the outcome for 95 participants treated with BPO and 49 participants treated with clindamycin. The risk ratio for achieving "much better" improvement on the scale was 1.77 (95% CI 0.82 to 3.81). These results consistently suggest similar effects of both treatments.

Draelos 2002, with 89 participants on BPO plus tazarotene and 87 participants on clindamycin plus tazarotene, assessed this outcome using a 5‐point Likert‐like scale (from highly favourable to highly unfavourable). At week 12, no significant difference was found in the proportion of participants rating improvement as highly favourable or favourable (92% versus 85%; P > 0.05). We cannot estimate RR as the effect size because data were not sufficient to calculate the number of participants with treatment success defined in our review.

Main pooled analyses: BPO versus clindamycin

Eight long‐term trials assessing this outcome suggest no clear differences in withdrawal due to adverse effects (Draelos 2002; Eichenfield 2011; Leyden 2001a; Lookingbill 1997; Study 156; Swinyer 1988; Thiboutot 2008; Tschen 2001) (RR 1.93, 95% CI 0.90 to 4.11; 3330 participants; 8 studies; I² = 0%; Analysis 3.2). Reasons for withdrawal from these trials included application site hypersensitivity, pruritus, erythema, face oedema, rash, and skin burning.

Subgroup analyses: co‐intervention

No evidence suggests that either BPO monotherapy or BPO add‐on treatment can increase the risk of withdrawal due to adverse events, with no significant differences between co‐intervention subgroups (P = 0.58).

Main pooled analyses: BPO versus clindamycin

The absolute change in TLs was ‐3.50 (95% CI ‐7.54 to 0.54; Analysis 3.3), derived from a total of 323 participants treated with BPO and 318 participants treated with clindamycin (Eichenfield 2011). This trial also showed the absolute change in ILs (‐1.20, 95% CI ‐2.99 to 0.59; Analysis 3.4) and NILs (‐2.40, 95% CI ‐5.30 to 0.50; Analysis 3.5). These results did not show significant differences in absolute change for each lesion type.

As for the percentage change, the mean difference reported in Lookingbill 1997 was in NILs (21.00, 95% CI 6.86 to 35.14; Analysis 3.7) but not in ILs (4.00, 95% CI ‐8.56 to 16.56; Analysis 3.6) (92 participants treated with BPO and 89 with clindamycin contributed to these two estimates).

Studies not included in the meta‐analysis

For the absolute change, Leyden 2001a reported the absolute change in both TLs (‐13.8 versus ‐6.2) and ILs (‐5.9 versus ‐3.2) in the BPO group (120 participants) compared to the clindamycin group (120 participants) at week 10. Tschen 2001 found some difference (‐12.4 in ILs for BPO (95 participants) versus ‐14.7 for clindamycin (49 participants); P < 0.034). In Tucker 1984, trial authors reported the absolute change in comedone, papule, pustule, and cyst lesions at week 10 separately: corresponding results were ‐6.8 versus ‐7.3, ‐9.5 versus ‐9.7, ‐5.7 versus ‐3.4, and ‐0.05 versus ‐2.3.

The percentage reduction in TLs, ILs, and NILs was 25.5% versus 23.5%, 33.5% versus 29.8%, and 18.8% versus 11.2% for the BPO group (n = 70) and the clindamycin group (n = 70) in Study 152. In Study 156, the percentage reduction in TLs, ILs, and NILs was 43.3% versus 33.3% (P = 0.008), 56.7% versus 48.6% (0.048), and 28.7% versus 18.0% (0.037) for the BPO group (n = 96) versus the clindamycin group (n = 96). Another trial found a significant difference in the percentage reduction in NILs (62.9% (BPO, 30 participants) versus 37.1% (clindamycin, 30 participants); P < 0.001) at week 12 (Swinyer 1988). For Thiboutot 2008, 809 participants were randomised to the BPO group and 812 to the clindamycin group. The percentage reduction was 41.6% versus 40.4% in TLs, 47.5% versus 46.2% in ILs, and 37.4% versus 36.2% in NILs at week 12. In Draelos 2002, the percentage reduction in both ILs (39.7% versus 40.2%; P > 0.05) and NILs (42.7% versus 55.4%; P > 0.05) at week 12 was similar between two groups. For the medium‐ and the short‐term period, both treatments reduced NILs by about 40% and 30%, respectively, with no significant differences. Similar effects on reducing ILs were also observed (31% versus 35% for the medium‐term period and 24% versus 22% for the short‐term period). We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Main pooled analyses: BPO versus clindamycin

The long‐term outcome was reported in two 12‐week trials (Eichenfield 2011; Thiboutot 2008). With a total of 1137 and 1140 participants treated with BPO and clindamycin, respectively, heterogeneity was substantial, with no significant difference in this outcome (RR 1.10, 95% CI 0.83 to 1.45; 2277 participants; 2 studies; I² = 72%; Analysis 3.8).

Main pooled analyses: BPO versus clindamycin

The risk of adverse events may be higher with BPO than with clindamycin (RR 1.24, 95% CI 0.97 to 1.58; 3018 participants; 6 studies; I² = 0%; Analysis 3.9). The most common adverse events reported in these trials included skin dryness, pruritus, burning, tingling, and headache.

Subgroup analyses: co‐intervention

No evidence suggests that co‐intervention affected the association between BPO and adverse events (P value for subgroup differences = 0.43; Analysis 3.9).

Main pooled analyses: BPO versus erythromycin

With 30 participants equally assigned to each group, an eight‐week trial reported that one participant in each group did not complete the trial due to adverse events (both had dermatitis) (Burke 1983). Evidence suggests that the risk of withdrawal was not different between the two groups (RR 1.00, 95% CI 0.07 to 15.26; 30 participants; Analysis 4.1).

Studies not included in the meta‐analysis

Of 44 participants treated with BPO and 45 with erythromycin, none withdrew due to adverse events during a 10‐week trial (Chalker 1983).

Studies not included in the meta‐analysis

After a six‐week treatment duration, reduction of 34.8% in NILs and 73.6% in ILs was found in the erythromycin group, which consisted of 25 participants; the counterpart in NILs and ILs was 70.1% and 63.9%, respectively, in the BPO group of 24 participants (Dogra 1993).

Mean lesion counts at each follow‐up were reported in another trial (Burke 1983). At the end of the trial, the mean difference in TLs, ILs, and NILs between BPO and erythromycin groups was ‐5.60 (95% CI ‐8.03 to ‐3.17), ‐0.90 (95% CI ‐1.72 to ‐0.08), and ‐8.40 (95% CI ‐9.87 to ‐6.93), respectively.

In one 16‐week parallel trial involving BPO plus adapalene versus erythromycin plus adapalene (only abstract was available), it is unclear how many participants were randomised to each group (Cassano 2002). With no details on the absolute or percentage reduction in lesion counts, study authors claimed that all these treatments similarly led to a significant reduction in the numbers of papules, pustules, and comedones.

Studies not included in the meta‐analysis

No adverse events were identified in one trial (Chalker 1983). In another trial, mild erythema and scaling were found in a small number of participants, but the exact number was not reported (Burke 1983). In Cassano 2002, tolerability was satisfactory in most cases in each treatment arm. Itching and burning occurred in the early phase in both groups, but the intensity of these symptoms was gradually decreased. We cannot calculate RR for these trials as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

Bissonnette 2009, a 12‐week trial, and Chantalat 2006, a six‐week trial, assessed this outcome; neither provided sufficient data to estimate the effect size (Appendix 9).

A total of 80 participants with mild to moderate facial acne were randomised to either BPO or salicylic acid, and no significant difference was found (P = 0.81) (Bissonnette 2009). According to the other trial, which included 21 participants treated with BPO and 20 with salicylic acid, self‐assessment results favoured salicylic acid in terms of improvements in tone and texture (Chantalat 2006).

Studies not included in the meta‐analysis

One trial reported that no participants terminated treatment due to adverse events (Appendix 10) (Bissonnette 2009).

Studies not included in the meta‐analysis

Evidence in Bissonnette 2009 showed no difference in ILs (P = 0.748) or NILs (P = 0.445) between two groups. With 30 participants randomised in a four‐week cross‐over trial, a significant reduction in comedones was found in those treated with salicylic acid cleanser (Shalita 1989). However, neither of the trial reports provided sufficient data to estimate the effect size.

Studies not included in the meta‐analysis

As reported in Chantalat 2006, participants treated with salicylic acid had a significant improvement in acne‐related quality of life (ARQL). This effect started at the second week and continued through the following four weeks. By contrast, such an effect was not found in the BPO group. Insufficient data were available from both trials to estimate the effect size.

Main pooled analyses: BPO versus salicylic acid

Although 16 of 80 participants in the trial had at least one adverse event during treatment (Bissonnette 2009), the number of participants with adverse events was not reported for each group. In Chantalat 2006, two of 21 participants treated with BPO experienced an adverse event, but no adverse events occurred among 20 participants in the salicylic acid group, with an RR of 4.77 (95% CI 0.24 to 93.67; Analysis 5.1).

Studies not included in the meta‐analysis

This outcome was reported only in a 12‐week trial comparing BPO (5%)/erythromycin (3%) with tretinoin (0.025%)/erythromycin (4%) among 112 participants (Gupta 2003), but the data provided were not sufficient to calculate the effect size of treatment success defined in our review. Participants rated the overall change in their facial acne using a 7‐point scale (‐1 = exacerbation, 0 = no change, 1 = modest clearing, 2 = marked change, 3 = good, 4 = almost cleared, 5 = completely cleared). Trial authors claimed that scores rated in the BPO/erythromycin group were lower than those in the tretinoin/erythromycin group, with a significant difference as early as week 2.

Main pooled analyses: BPO versus tretinoin

Three trials reported the long‐term outcome, with no statistical difference between BPO and tretinoin (RR 1.11, 95% CI 0.07 to 17.36; 200 participants; I² = 0%; Analysis 6.1) (Bowman 2005; Gupta 2003; Jackson 2010). In one 10‐week trial (Bowman 2005), in which 43 were treated with BPO/clindamycin and 45 with tretinoin plus clindamycin, no participants in each group discontinued treatments due to adverse events during the trial. In one 16‐week trial (Jackson 2010), which included 27 participants in each group (BPO/clindamycin versus tretinoin/clindamycin), no participants discontinued treatment because of adverse effects. Only two participants (2/112) in Gupta 2003 did not complete the study due to severe burning or stinging, with one in each group.

Studies not included in the meta‐analysis

We cannot estimate MD as the effect size for this outcome because data were not sufficient to calculate standard deviation as reported in the included studies. In an eight‐week trial analysing 54 participants treated with BPO acetone gel (Lyons 1978), 38 treated with BPO alcohol/detergent gel and 55 with tretinoin cream, trial authors reported that no statistical differences in the change in comedones, pustules, and total lesions were found between groups. However, both BPO gels reduced papules by 59% ‐ significantly higher than 29% with tretinoin cream.

In one 12‐week parallel trial with 33 participants randomised to each group (BPO plus clindamycin versus tretinoin plus clindamycin) (Kaur 2015), the mean number of ILs was reduced from 3.6 to 0 among participants on BPO plus clindamycin and from 4.9 to 0.3 among participants on tretinoin plus clindamycin. The mean number of NILs was reduced from 12.0 to 0 among participants on BPO plus clindamycin, and from 13.7 to 1.3 among participants on tretinoin plus clindamycin. It is unclear whether there was a significant difference between the two groups.

In one 10‐week trial (Bowman 2005), the percentage reduction in TLs, ILs, and NILs at week 10 in the BPO/clindamycin group was 61.4%, 65.7%, and 57.2%, respectively, compared to 49.6%, 52.5%, and 46.2% in the other group. A significant difference was found in the percentage reduction in ILs (P = 0.02). For medium‐term and short‐term treatment durations, the percentage reduction in TLs (medium term: 51.2% versus 40.9%; short term: 42.1% versus 35.3%), ILs (61.0% versus 44.9%; 53.9% versus 45.7%), and NILs (42.8% versus 37.9%; 32.0% versus 21.9%) was consistently greater in the BPO/clindamycin treatment group, but it is unclear whether significant differences existed, as no relevant information was reported.

The median percentage reduction reported in Jackson 2010 was 52.4% versus 54.3% in TLs, 74.1% versus 70.7% in ILs, and 53.3% versus 52.8% in NILs at week 16. The median percentage reduction in TLs was 52.4% versus 54.3%, 51.8% versus 44.5%, and 43.8% versus 31.3% for the long‐, medium‐, and short‐term periods. Trial authors claimed that the difference between groups was comparable, except that the short‐term period favoured BPO/clindamycin treatment. No significant difference was observed in the median percentage reduction in ILs and NILs at any time points (long term: 74.1% versus 70.7% in ILs, 53.3% versus 52.8% in NILs; medium term: 61.6% versus 46.7% in ILs, 30.9% versus 36.8% in NILs; short term: 53.7% versus 32.5% in ILs, 39.9% versus 25.1% in NILs).

No significant difference was found in Gupta 2003 for the absolute reduction in ILs or NILs at any time points between groups. The mean number of papules was reduced from 16.0 at baseline to 10 (short term), to 8 (medium term), and to 5.5 (long term) among participants on BPO/erythromycin The counterpart in participants on tretinoin/erythromycin was reduced from 17.7 to 12, to 8, and to 5.5, respectively. The number of pustules in BPO/erythromycin was reduced from 7.1 to 3.7, to 3, and to 2.3 for short‐, medium‐, and long‐term periods, respectively, and the counterpart in tretinoin/erythromycin was decreased from 8.3 to 3.7, to 2, and to 1.5. The mean number of comedones was reduced from 28.2 to 21.3, to 16.3, and to 9.3 in participants on BPO/erythromycin for short‐, medium‐, and long‐term periods, respectively. The counterpart was reduced from 35.5 to 21.3, to 13.1, to 11.7 in participants on tretinoin/erythromycin.

Main pooled analyses: BPO versus tretinoin

At week 10, the percentage of participants rated 'clear' or 'almost clear' on the IGA scale of acne severity was not significantly different between the BPO/clindamycin group and the tretinoin plus clindamycin group (RR 2.09, 95% CI 0.86 to 5.08; 88 participants; Analysis 6.2).

Studies not included in the meta‐analysis

This outcome was reported only in Jackson 2010 for this comparison. The reduction in total, erythromycin‐resistant, and clindamycin‐resistant C acnes strains at week 16 (log₁₀ CFU cm^‐2) was ‐1.84 versus ‐0.78 (P < 0.003), ‐0.83 versus 0.3 (P < 0.002), and ‐0.73 versus 0.05 (P < 0.002).

Main pooled analyses: BPO versus tretinoin

For the long‐term treatment duration, we did not observe significant differences between the two groups, with an RR of 0.58 (95% CI 0.31 to 1.07; 166 participants; 2 studies; I² = 29%; Analysis 6.3). Trial authors did not report further information regarding what adverse events were observed in Jackson 2010. The most frequently reported adverse events in Gupta 2003 were facial dryness, stinging/burning, erythema, pruritus, and peeling/scaling. No significant differences were found between groups for any individual adverse events, except erythema (6% versus 19%; P = 0.047).

The outcome reported in trials for BPO monotherapy versus tretinoin was only for a medium‐term treatment duration (103 participants in BPO groups and 67 in the tretinoin group), with an RR of 0.03 (95% CI 0.00 to 0.48; 170 participants; 2 studies; I² = 0%; Analysis 6.4). Shahid 1996 specified that no participants in both groups reported any adverse events. In Lyons 1978, no adverse events were reported in either of the two BPO groups, but 18.2% (10/55) of participants treated with tretinoin cream experienced at least one adverse event, including burning, erythema, peeling, and soreness.

Subgroup analyses: co‐intervention

No evidence suggests that this association may differ by co‐intervention (P value for subgroup differences = 0.24; Analysis 6.3).

Studies not included in the meta‐analysis

We cannot calculate RR as the effect size for these trials because the number of participants with at least one adverse event was not provided (Bowman 2005; Kaur 2015).

Burning sensation and dryness were the only two adverse events that occurred during the Kaur 2015. Burning sensation occurred in one participant on BPO plus clindamycin, and dryness occurred in four participants on tretinoin plus clindamycin.

Although no significance test results were provided, Bowman 2005 reported that adverse events were substantially more common among participants on BPO/clindamycin than among those on tretinoin and clindamycin. The most common adverse events were irritation (11.6% versus 17.9%), dryness (7.0% versus 15.6%), desquamation (7.0% versus 11.1%), and erythema (4.7% versus 8.9%).

Studies not included in the meta‐analysis

In Marazzi 2002, participants rated the overall change in their facial acne as improved, no change, or worse. There was no significant difference between the two groups in the proportion of participants rated as "improved" (P value = 0.26).

Main pooled analyses: BPO versus isotretinoin

The risk of withdrawal due to adverse effects was not significantly different (RR 1.23, 95% CI 0.53 to 2.87; 239 participants; 2 studies; I² = 0%; Analysis 7.1) in Hughes 1992 and Marazzi 2002). Hughes 1992 did not specify what adverse events led to their discontinuation. Nine participants on BPO/erythromycin and eight participants on isotretinoin/erythromycin withdrew due to adverse events, with no significant differences between the two groups (Marazzi 2002). The major event leading to withdrawal was infection requiring antibiotic treatment. Besides, two participants on BPO/erythromycin and five on isotretinoin/erythromycin withdrew because of skin reactions to these products.

Subgroup analyses: co‐intervention

No evidence suggests that this association may differ by co‐intervention (P value for subgroup differences = 0.24; Analysis 7.1).

Studies not included in the meta‐analysis

No participants in Cunliffe 2001 withdrew as a result of any adverse events.

Main pooled analyses: BPO versus isotretinoin

For the comparison between BPO/erythromycin and isotretinoin/erythromycin, no significant difference was found in the absolute reduction in ILs (MD ‐4.00, 95% CI ‐9.89 to 1.89; Analysis 7.2) for the long‐term period. However, the counterpart for medium‐term and short‐term periods was significantly different, favouring BPO/erythromycin, with an MD of ‐6.10 (95% CI ‐11.27 to ‐0.93; Analysis 7.4) and ‐9.20 (95% CI ‐14.09 to ‐4.31; Analysis 7.6), respectively. In terms of the absolute change in NILs, no significant difference was observed for any time periods (long term: MD 2.30, 95% CI ‐6.07 to 10.67; Analysis 7.3; medium term: MD 4.90, 95% CI ‐1.66 to 11.46; Analysis 7.5; short‐term: MD 1.20, 95% CI ‐4.39 to 6.79; Analysis 7.7).

Studies not included in the meta‐analysis

The outcome of lesion counts, rather than the change in lesion counts, was reported in Hughes 1992. At the end of the trial (12 weeks), the mean difference between BPO and isotretinoin groups was ‐6.30 (95% CI ‐17.04 to 4.44). The mean difference for the medium‐term outcome (eight weeks) was ‐0.60 (95% CI ‐10.70 to 9.50). This outcome was not available for Cunliffe 2001.

Main pooled analyses: BPO versus isotretinoin

This outcome was reported in two trials (Cunliffe 2001; Marazzi 2002).

In Marazzi 2002, 53 participants in the BPO/erythromycin group and 64 participants in the isotretinoin/erythromycin group experienced at least one adverse event during the long‐term period, with no significant differences (RR 0.85, 95% CI 0.68 to 1.06; 188 participants; Analysis 7.8). The most common treatment‐related events were burning reported for three participants using BPO/erythromycin and five participants using isotretinoin/erythromycin, followed by itching or redness reported in three participants using BPO/erythromycin and four participants using isotretinoin/erythromycin.

As reported in Cunliffe 2001, the risk ratio for the short‐term outcome was 0.18 (95% CI 0.05 to 0.69; Analysis 7.9) in favour of BPO treatment.

Studies not included in the meta‐analysis

Instead of presenting the total number, authors of the Hughes 1992 trial described only the numbers of participants for individual adverse events. The most common adverse event for both groups was redness (10/26 for the BPO group and 10/25 for the isotretinoin group), followed by scaling (7/26 and 7/25, respectively).

Studies not included in the meta‐analysis

No withdrawal due to adverse events was reported in Stinco 2007.

Studies not included in the meta‐analysis

At the end of Stinco 2007, ILs were reduced by 45% in the azelaic acid group and 44% in the BPO group. NILs were reduced by 47% for azelaic acid and 38% for BPO. Trial authors reported no significant differences between treatments.

Studies not included in the meta‐analysis

Although no data on the total number of participants experiencing an adverse event were provided in Stinco 2007, trial authors reported individual adverse events. For the BPO group, 12 participants reported itchiness (seven mild and five moderate) after two weeks of treatment. For the azelaic acid group, 20 participants reported itchiness (12 mild, five moderate, and three severe). Dryness occurred in 15 participants treated with BPO (one mild, nine moderate, and five severe) and in 14 treated with azelaic acid (nine mild, three moderate, and two severe). A total of 10 participants in the BPO group (eight mild and two moderate) and six in the azelaic acid group (five mild and one moderate) experienced erythema.

Studies not included in the meta‐analysis

After six‐week treatment duration, although reduction in NILs and ILs was 75.2% and 60.7%, respectively, in the retinoic acid group, the counterpart in the BPO group was 70.1% and 63.9% (Dogra 1993).

Studies not included in the meta‐analysis

This trial assessed the outcome using a 4‐point Likert‐like scale (from worse = ‐1 to greatly improved = 2). The average score was 0.66 and 0.75 for the BPO and sulphur groups, respectively.

Studies not included in the meta‐analysis

The change in lesion counts was assessed for superficial (comedones and pustules) and deep (papules and cysts) lesions via an 11‐point scoring system (decrease in lesions by 100% = 5, 76% to 99% = 4, 51% to 75% = 3, 26% to 50% = 2, < 25% = 1, no change = 0; increase by < 25% = ‐1, 26% to 50% = ‐2, 51% to 75% = ‐3, 76% to 100% = ‐4, over 100% = ‐5). The average score for superficial lesions was 0.55 and ‐0.70 for BPO and sulphur groups, respectively, and the counterpart for deep lesions was 0.69 and 0.30.

Studies not included in the meta‐analysis

The study provided only information on individual adverse events. Excessive erythema and dryness occurred in five participants on BPO and four on sulphur. One participant developed severe allergic sensitisation after BPO treatment. We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

One trial including a total of 120 participants assessed this outcome (Tung 2014). However, the data, which were available in an abstract, were not sufficient to estimate the effect size. Study authors reported that both acne regimens were effective but found no statistical difference across arms (P = 0.70).

Studies not included in the meta‐analysis

Insufficient data were available to estimate the effect size of either absolute or percentage change. Milani 2003, which included 30 participants treated with BPO and 30 with hydrogen peroxide, reported the mean counts of lesions at week 8. The mean difference in TLs, ILs, and NILs was ‐2.00 (95% CI ‐6.08 to 2.08), 0.00 (95% CI ‐1.52 to 1.52), and ‐2.00 (95% CI ‐4.09 to 0.09), respectively, with no significant difference.

Capizzi 2004, a parallel trial with 26 participants in each group (BPO plus adapalene versus hydrogen peroxide plus adapalene), suggested that BPO reduced the mean number of TLs from 40 to 5.4 and hydrogen peroxide from 44 to 3.2. ILs were decreased from 21 to 3.1 among participants on BPO, and from 25 to 2.5 among participants on hydrogen peroxide, with NILs decreasing from 19 to 2.3 and from 19 to 0.7 (P = 0.0025), respectively.

Main pooled analyses: BPO versus isotretinoin

The outcome was reported in both trials. In Milani 2003, seven of 30 participants with BPO and two of 30 participants with hydrogen peroxide had at least one adverse event, with no significant difference between the two groups (RR 3.50, 95% CI 0.79 to 15.49; 60 participants; 1 study; Analysis 8.1).

Studies not included in the meta‐analysis

No adverse events were found in each group in the other trial (Tung 2014), where the sample size for each group was not reported. We cannot calculate RR as the effect size for Capizzi 2004 because the number of participants with at least one adverse event was not provided. Dryness, erythema, and burning were observed (19/26 versus 2/26, 5/26 versus 1/26, 4/26 versus 0/26 for BPO and hydrogen peroxide, respectively). A significant difference was found in the occurrence of dryness (P = 0.0025) and burning (P = 0.01).

Studies not included in the meta‐analysis

No withdrawal due to adverse events was reported in that trial.

Studies not included in the meta‐analysis

At the end of the trial, the mean number of ILs was reduced from 35 to 19 in the BPO group and from 34 to 12 in the superoxidised solution group. The proportion of participants with at least 50% reduction in ILs was 17 of 24 versus 30 of 39 with no significant difference.

Studies not included in the meta‐analysis

Trial authors reported that no local adverse events were noted in each group. We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

Although no severe adverse events were observed in the group receiving treatment with Casuarina equisetifolia bark extract, 17% of participants on BPO treatment complained of irritation, redness, and swelling.

Studies not included in the meta‐analysis

This outcome was assessed using a questionnaire (no details about the questionnaire were specified) only for the Chinese herbal mask group and not for the BPO group. Of 111 respondents in the assessment, 104 (93.7%) were very satisfied or satisfied with the treatment.

Studies not included in the meta‐analysis

A total of 3 of 120 participants on BPO treatment and 2 of 112 participants on Chinese herbal mask complained of irritation and redness. We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

This outcome was assessed in the trial, but it is unclear how the outcome was assessed. It was reported that participants rating treatments as "good to excellent" accounted for 63% and 49% in the two groups, respectively, with an RR of 1.32 (95% CI 0.86 to 2.02).

Primary outcomes: withdrawal due to adverse effects

Main pooled analyses: BPO versus meclocycline

It was reported that no participants discontinued meclocycline treatment due to adverse effects, but three discontinued BPO treatment for this reason, with an RR of 6.43 (95% CI 0.34 to 120.03; Analysis 10.1), Adverse events leading to withdrawal were not specified.

Studies not included in the meta‐analysis

For long‐term outcomes, the difference in absolute reduction in papules, pustules, and comedones was ‐5.3, 1.1, and ‐2.7, respectively. For medium‐term outcomes, the counterpart was ‐4.7, ‐0.1, and ‐3, respectively. For short‐term outcomes, the counterpart was ‐6.7, ‐0.6, and ‐1.9, respectively. All results were statistically significant (P < 0.05) except for the reduction in pustules for long‐term data.

Studies not included in the meta‐analysis

Trial authors reported that it was more common for BPO‐treated participants to have local side effects (including erythema, scaling, stinging/burning) than meclocycline‐treated participants. Yellowish decolouration was reported in one participant on BPO and in two participants on meclocycline. Allergic contact dermatitis was confirmed in one participant treated with BPO. We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

At week 12, BPO reduced the mean number of ILs from 33.3 to 12.7, and isolutrol from 23.9 to 15.7. NILs were decreased from 25.5 to 11.9 among participants on BPO and from 23.4 to 16.4 among participants on isolutrol. Trial authors presented significance test results for before‐and‐after lesion counts in each group rather than differences between the two groups.

Main pooled analyses: BPO versus isolutrol

Overall, adverse events occurred in 94% of BPO‐treated participants, whereas only 34% of isolutrol‐treated participants reported such problems, with higher risk in the BPO group (RR 2.75, 95% CI 1.73 to 4.38; 70 participants; Analysis 9.1). The most common event reported in the BPO group was dryness (83%), followed by erythema (49%). Dryness (31%) and pruritus (14%) were the most common events in the isolutrol group.

Studies not included in the meta‐analysis

Insufficient data were available in this trial to estimate the effect size of absolute or percentage change. It was reported that the mean difference in absolute reduction in ILs was ‐10.4 (P < 0.001), ‐10.8 (P < 0.001), and ‐10.2 (P < 0.05) for the long, medium, and short term, respectively. The counterpart in NILs was ‐1.2, ‐1.0, and ‐2.3, respectively (all P values > 0.05).

Main pooled analyses: BPO versus tea tree oil

Overall, 79% (50/63) of BPO‐treated participants and 44% (27/61) of tea tree oil‐treated participants reported adverse events during the trial, with an RR of 1.79 (95% CI 1.32 to 2.44; Analysis 11.1). The most common adverse events reported included skin dryness, pruritus, stinging, burning, and redness.

Studies not included in the meta‐analysis

Four participants in each group did not complete the study; however, none of these participants withdrew from the trial due to adverse effects.

Studies not included in the meta‐analysis

Trial authors presented significance test results for before‐and‐after lesion counts in each group rather than the difference between the two groups based on the Global Acne Severity Scale (Tabasum 2014). At week 6, grading scores for comedones and papules were significantly different before and after both treatments (both P values < 0.001). Although significant results were found for the grading scores for pustules (P < 0.001) and nodules (P = 0.02) in the Unani preparation group, scores in the BPO group were not significantly reduced from baseline.

Studies not included in the meta‐analysis

Similar to the outcome mentioned above, trial authors presented significance test results for before‐and‐after quality of life in each group rather than the difference between two groups. The mean score of the Cardiff Acne Disability Index was reduced from 12.00 to 5.80 in the BPO group (P < 0.001) and from 12.25 to 4.95 in the Unani preparation group (P < 0.001).

Studies not included in the meta‐analysis

Trial authors claimed that the difference between two groups regarding adverse events was not significant. The most common events reported in both groups were itching (9/20 versus 13/20) and dryness (9/20 versus 7/20). We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

After a six‐week treatment duration, although a reduction in NILs and 1ILs was 7.4% and 11.9% in the glycerin group of 21 participants, respectively, the counterpart was 70.1% and 63.9% in the BPO group of 24 participants (Dogra 1993).

Studies not included in the meta‐analysis

At week 12, both groups showed marked improvement in both ILs (60% versus 62%) and NILs (54% versus 56%). With no statistically significant difference in such reduction between groups, trial authors concluded that both treatments were similarly effective in reducing lesions.

Studies not included in the meta‐analysis

Trial authors mentioned that erythema and photosensitivity were significantly less in the chloroxylenol/salicylic acid group at week 12 than in the BPO group (P = 0.0002 and P = 0.05, respectively). We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Main pooled analyses: BPO versus chloroxylenol/zinc oxide

The medium‐term outcome was assessed according to the percentage of improvement (as worse, no change (0 to 9%), mild improvement (10% to 39%), moderate improvement (40% to 69%), marked improvement (70% to 89%), and clearance (> 90%)), with no significant difference in treatment success (RR 0.89, 95% CI 0.51 to 1.56; 26 participants; Analysis 12.1).

Studies not included in the meta‐analysis

Two participants discontinued treatment due to flare‐ups. However, it is unclear what treatments they received before the flare‐ups.

Main pooled analyses: BPO versus chloroxylenol/zinc oxide

This trial reported the percentage change in ILs and NILs. Compared with the chloroxylenol/zinc oxide group, mean differences in the percentage change were ‐7.70% (95% CI ‐19.36% to 3.96%) and ‐5.60% (95% CI ‐14.74% to 3.54%), respectively.

Studies not included in the meta‐analysis

No significant difference in the number of flare‐ups was found (2/13 on BPO versus 1/13 on chloroxylenol/zinc oxide). Dryness and/or peeling of the skin was reported by four participants on BPO and one participant on chloroxylenol/zinc oxide (P < 0.01). We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Main pooled analyses: BPO/adapalene versus placebo or no treatment

Overall, there was no significant difference in long‐term treatment success between the two groups (RR 1.16, 95% CI 0.97 to 1.38; 1480 participants; 3 studies; I² = 84%; Analysis 13.1).

Subgroup analyses: co‐intervention

Effects differed by the co‐intervention (P value for subgroup differences = 0.02), with a larger effect size in the BPO/adapalene versus the placebo or no treatment group (RR 1.25, 95% CI 1.09 to 1.44) than with the co‐intervention of lymecycline (RR 0.99, 95% CI 0.87 to 1.13) (Analysis 13.1).

Studies not included in the meta‐analysis

We did not pool results from the other four trials assessing this outcome because the number of treatment successes defined in our review was not available (Eichenfield 2013; Gold 2010; Gollnick 2009; Thiboutot 2007).

Eichenfield 2013 randomised 142 participants to the BPO/adapalene group and 143 to the placebo group and assessed the outcome using a 6‐point Likert‐like scale. The risk ratio for self‐reported complete or marked improvement was 2.93 (95% CI 2.14 to 4.01).

In Gollnick 2009, 419 participants in the BPO/adapalene group and 418 in the placebo group were assessed via a 6‐point Likert‐like scale. The risk ratio for self‐reported complete or marked improvement was 1.75 (95% CI 1.46 to 2.09).

A similar 6‐point Likert‐like scale was used in another trial (Thiboutot 2007), in which 149 participants were treated with BPO/adapalene and 71 participants with placebo. The risk ratio for self‐reported complete or marked improvement was 3.00 (95% CI 1.64 to 5.49).

Gold 2010, a 12‐week trial including 232 participants in the BPO/adapalene plus doxycycline group and 227 in the placebo plus doxycycline group, assessed participant satisfaction at week 12 and found significant differences in the proportion of participants reporting "very satisfied" or "satisfied" (scale used for the assessment was not specified) in favour of the BPO/adapalene add‐on treatment group (76.3% versus 50.3%; P < 0.001).

These results consistently suggested a significant difference between the two groups in favour of BPO/adapalene. Sensitivity analysis additionally including these four trials suggests a higher probability of self‐assessed improvement in the BPO/adapalene group, with an RR of 1.55 (95% CI 1.25 to 1.91).

Main pooled analyses: BPO/adapalene versus placebo or no treatment

The risk ratio of withdrawal due to adverse effects was 2.28 (95% CI 1.10 to 4.71; 3801 participants; I² = 3%; Analysis 13.2), pooled from the eight parallel‐group trials with a total of 2199 participants on BPO/adapalene and 1602 on placebo (Dreno 2011; Eichenfield 2013; Gold 2009; Gold 2010; Gold 2016; Gollnick 2009; Thiboutot 2007; Weiss 2015), suggesting higher risk of discontinuation in the BPO/adapalene group. The adverse events reported leading to withdrawal included erythema, irritation, and atopic dermatitis flare.

Subgroup analyses: co‐intervention

No evidence suggests that the association may differ by co‐intervention (P value for subgroup differences = 0.12) (Analysis 13.2).

Studies not included in the meta‐analysis

Two split‐face design trials reported that one of 38 participants discontinued BPO/adapalene because of moderate skin irritation (Dreno 2016), and that two of 67 participants had treatment‐related adverse events on the BPO/adapalene side (one with moderate pain of skin and one with mild skin irritation) that led to study discontinuation (Dréno 2018), respectively.

Studies not included in the meta‐analysis

In the split‐face trial (Dréno 2018), the percentage reduction in acne lesion counts at week 24 was greater for the BPO/adapalene side than for the placebo side, with differences of 22.4% (73.3 versus 50.9%), 28.8% (86.7 versus 57.9%), and 18.1% (59.5 versus 41.4%) for total, inflammatory, and non‐inflammatory lesions, respectively.

Eichenfield 2013 found a significant difference in percentage reduction in TLs (68.6% versus 19.3%), ILs (63.2% versus 14.3%), and NILs (70.7% versus 14.6%) (all P values < 0.001).

Similar results were found in Gold 2009, showing a significant difference in percentage reduction in TLs (60.7% versus 31.7%), ILs (66.0% versus 36.9%), and NILs (58.9% versus 32.9%) between groups (all P values < 0.05).

In Gollnick 2009, compared with placebo groups, the percentage reductions in TLs, ILs, and NILs were significantly higher in the BPO/adapalene group (68.8% versus 41.1%, 72.4% versus 49.8%, and 66.4% versus 41.9%, respectively) (all P values < 0.05).

Thiboutot 2007 found 51.0% and 31% reduction in TLs, 62.9% and 37.8% in ILs, and 51.2% and 37.5% in NILs for BPO/adapalene and adapalene, respectively (all P values < 0.001).

In Weiss 2015, which compared BPO/adapalene gel with placebo among participants with moderate and severe acne (217 versus 69 participants), the absolute changes in ILs for each group were ‐27.0 and ‐14.4, respectively. The absolute changes in NILs were ‐40.1 and ‐18.4, respectively. This trial also reported percentage changes in ILs (‐68.7% vs. ‐39.2%) and NILs (‐68.3% vs. ‐37.3%, respectively). Trial authors claimed the P values for all changes to be less than 0.001.

As reported in Dreno 2011, a significant difference was found in the percentage reduction in TLs (74.1% versus 56.8%; P < 0.001), ILs (81.7% versus 71.0%; P < 0.001), and NILs (71.7% versus 52.5%; P < 0.001) between the BPO/adapalene group and the placebo group.

Main pooled analyses: BPO/adapalene versus placebo or no treatment

Pooled results for the long‐term outcome support the use of BPO/adapalene (RR 2.43, 95% CI 1.80 to 3.28; 3012 participants; 6 studies; I² = 77%; Analysis 13.3). Three studies focusing on BPO/adapalene used alone reported medium‐ and short‐term outcomes, with an RR of 2.56 (95% CI 1.88 to 3.47; 2175 participants; I² = 31%; Analysis 13.4) and an RR of 2.76 (95% CI 1.84 to 4.16; 2175 participants; I² = 0%; Analysis 13.5), respectively.

Subgroup analyses: co‐intervention

For the long‐term outcome, effects of BPO differed by the co‐intervention (P value for subgroup differences < 0.0001; Analysis 13.3). For BPO/adapalene used alone, the pooled risk ratio for the long‐term outcome was 2.45 (95% CI 2.07 to 2.90; I² = 1%) (Eichenfield 2013; Gold 2009; Gollnick 2009; Thiboutot 2007). A significant difference was found for this outcome (74/232 versus 19/227) favouring BPO/adapalene plus doxycycline treatment (RR 3.81, 95% CI 2.38 to 6.10) in Gold 2010. Dreno 2011 showed a significant difference in this outcome (91/191 versus 63/187), with an RR of 1.41 (95% CI 1.10 to 1.82).

Studies not included in the meta‐analysis

In a split‐face trial (Dréno 2018), 'clear' or 'almost clear' was 45% more likely to occur with BPO/adapalene than with placebo (64.2 versus 19.4%).

Main pooled analyses: BPO/adapalene versus placebo or no treatment

Overall, BPO/adapalene tended to increase the long‐term risk of adverse events (RR 2.67, 95% CI 1.15 to 6.19; 2465 participants; 6 studies; I² = 86%; Analysis 13.6). The most common adverse events reported in these trials included burning sensation, skin irritation, dryness, and headache.

Subgroup analyses: co‐intervention

Substantial heterogeneity may be due to the co‐interventions (test for subgroup differences: P = 0.0002; Analysis 13.6). For BPO/adapalene used alone, adverse events were assessed in all five trials, four of which presented sufficient data for pooling, including 927 participants in the BPO/adapalene group and 701 in the placebo group (Eichenfield 2013; Gollnick 2009; Thiboutot 2007; Weiss 2015). The pooled RR was 4.71 (95% CI 2.36 to 9.37; I² = 36%; Analysis 13.6), suggesting the risk of adverse events was significantly higher in the BPO/adapalene treatment group. As for BPO/adapalene add‐on treatment, however, we did not find increased risk (Dreno 2011; Gold 2010).

Studies not included in the meta‐analysis

Gold 2009 did not present the number of participants experiencing at least one adverse event but reported that the number was similar across study treatments. Dréno 2018, a split‐face trial, showed that treatment‐related adverse events were reported in 25.4% of participants, with 20.9% on the BPO/adapalene side versus 9.0% on the placebo side. The most common were skin irritation (14.9% versus 6.0%) and skin pain (3.0% versus 1.5%).

Studies not included in the meta‐analysis

Rather than summarising the number of adverse events, trial authors reported local side effects individually during the three‐week period. The number of participants experiencing erythema, scaling, dryness, and stinging/burning was 25 versus 28, 25 versus 18, 34 versus 28, and 51 versus 57, respectively. We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Studies not included in the meta‐analysis

The proportion of participants rating BPO plus adapalene as better, similar, or worse than salicylic acid was 25.8%, 41.9%, and 32.2%, respectively (Zheng 2019).

Studies not included in the meta‐analysis

During the trial, 3 out of 34 participants developed irritant contact dermatitis on BPO treatment and withdrew from the study (Zheng 2019).

Studies not included in the meta‐analysis

Trial authors reported no significant difference in reducing ILs and NILs between groups (Makino 2015). After four weeks, BPO plus adapalene showed similar reduction in papules/pustules (49.8% versus 47.9%), non‐inflammatory lesions (42.7% versus 43.1%), and total lesions (45.6% versus 44.1%) (Zheng 2019).

Studies not included in the meta‐analysis

Trial authors reported no significant difference in the mean scores of IGA between groups (Makino 2015).

Studies not included in the meta‐analysis

No participants developed adverse events, except for three, who discontinued treatment due to irritant contact dermatitis at the BPO treatment site (Zheng 2019).

Studies not included in the meta‐analysis

Trial authors claimed no significant change in lesion counts between groups. However, no relevant data were presented.

Studies not included in the meta‐analysis

Trial authors did not present the number of adverse events but the mean score of erythema, dryness/scaling, itching, and burning/stinging on a 4‐point scale (from 0 = none to 3 = severe). Mean differences were 0.4 (P < 0.001), 0.201 (P < 0.001), 0.01 (P = 0.162), and 0.076 (P = 0.401), respectively. We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided.

Main pooled analyses: BPO/clindamycin versus placebo or no treatment

This outcome was assessed in five trials (Leyden 2001a; Pariser 2014; Tanghetti 2006; Thiboutot 2008; Tschen 2001), only one of which presented data in accordance with the definition of the outcome in our review (Leyden 2001a). Results from these trials consistently showed significant difference between the two groups.

Leyden 2001a included 120 participants in the BPO/clindamycin group and 120 in the placebo group. The outcome was assessed with a 5‐point Likert‐like scale. The risk ratio for treatment success was 2.95 (95% CI 1.96 to 4.46; Analysis 14.1) with significant differences between the two groups.

Studies not included in the meta‐analysis

We cannot pool this result with those from the other four trials because they did not present the number of treatment successes as defined in our review (Pariser 2014; Tanghetti 2006; Thiboutot 2008; Tschen 2001). Pariser 2014 randomised 253 participants to the BPO/clindamycin group and 245 to the placebo group and assessed the outcome using a 7‐point Likert‐like scale. At week 12, the risk ratio for self‐reported ‘clear’ or ‘almost clear’ was 1.98 (95% CI 1.44 to 2.73). A similar 7‐point Likert‐like scale was used in Thiboutot 2008, in which 797 participants in the BPO/clindamycin group and 395 participants in the placebo group were assessed. The risk ratio for self‐reported ‘clear’ or ‘almost clear’ was 2.34 (95% CI 1.85 to 2.97). Another 7‐point Likert‐like scale was used in Tschen 2001, in which 95 participants were treated with BPO/clindamycin and 48 with placebo. The risk ratio for self‐reported ‘much better’ was 4.04 (95% CI 1.52 to 10.77). In one 12‐week trial including 60 participants on BPO/clindamycin and 61 on placebo (Tanghetti 2006), all of whom received tazarotene as a co‐intervention, participants rated their overall impression of their treatment as highly favourable, favourable, unfavourable, or highly unfavourable. There was no significant difference in the proportion of participants rating improvement as highly favourable or favourable between the two groups (94% versus 90%) at week 12. Sensitivity analysis including these trials showed an RR of 2.10 (95% CI 1.25 to 3.52).

Main pooled analyses: BPO/clindamycin versus placebo or no treatment

No evidence suggests that participants treated with BPO/clindamycin had a higher risk of withdrawal due to adverse effects, with an RR of 1.07 (95% CI 0.38 to 3.00; 3095 participants; 8 studies; I² = 0%; Analysis 14.2). The most common adverse events leading to withdrawal included erythema, face oedema, rash, pruritus, and skin burning.

Subgroup analyses: co‐intervention

We did not observe significant differences between different co‐intervention subgroups (test for subgroup differences: P = 0.28; Analysis 14.2).

Main pooled analyses: BPO/clindamycin versus placebo or no treatment

For the mean change in lesion counts, Eichenfield 2011, which included 322 participants treated with BPO/clindamycin and 329 treated with placebo, found significant differences in the reduction in TLs (MD ‐15.20; 95% CI ‐19.57 to ‐10.83; Analysis 14.3), ILs (MD ‐5.10; 95% CI ‐6.83 to ‐3.37; Analysis 14.4), and NILs (MD ‐10.00; 95% CI ‐13.20 to ‐6.80; Analysis 14.5).

For the percentage change in lesion counts, we pooled two trials including 944 and 485 participants treated with BPO/clindamycin and placebo, respectively (Lookingbill 1997; NCT01138514). The mean difference in percentage reduction in ILs and NILs was 44.16% (95% CI 23.53 to 64.79%; 2 studies; I² = 86%; Analysis 14.6) and 37.65% (95% CI 26.04 to 49.25%; 2 studies; I² = 50%; Analysis 14.7), respectively, with substantial heterogeneity. The two trials were different in acne severity (Lookingbill 1997 covered mild acne, but NCT01138514 exclusively focused on moderate to severe acne).

Studies not included in the meta‐analysis

For the following trials, we cannot estimate MD as the effect size for this outcome because data were not sufficient to calculate standard deviation.

Regarding the mean change in lesion counts, Leyden 2001a claimed that both TLs (‐18.4 versus ‐1.3) and ILs (‐10.1 versus 0.7) were significantly reduced in the BPO/clindamycin group when compared to the placebo group at week 10. Tschen 2001 found ‐16.6 versus ‐11.3 in ILs for BPO/clindamycin (95 participants) and placebo (48 participants) (P < 0.034).

As for the percentage change in lesion counts, trial data published on the registry show 63.8% reduction in ILs among 218 participants treated with BPO/clindamycin and 49.9% among 185 participants treated with placebo (NCT01044264).

In Pariser 2014, BPO/clindamycin was statistically superior to placebo in terms of lesion percentage reduction in both ILs and NILs. The percent change in ILs from baseline to week 12 was 60.6% with BPO/clindamycin versus 31.4% with placebo, and 51.6% versus 27.4% in NILs (both P values < 0.001).

Similar results were found in Thiboutot 2008, in which 797 participants were randomised to the BPO/clindamycin group and 395 to the placebo group. The percentage reduction was 47.9% versus 26.2% in TLs, 54.6% versus 29.0% in ILs, and 43.2% versus 24.0% in NILs at week 12 (all P values < 0.001).

In Study 152, the percentage reduction in TLs, ILs, and NILs was 32.5% versus 20.6% (P = 0.003), 43.4% versus 28.6% (0.046), and 25.7% versus 15.4% (0.008) for the BPO/clindamycin group (n = 73) and the placebo group (n = 37).

Del 2007, a 12‐week parallel trial including 37 participants on BPO/clindamycin plus adapalene and 37 participants on adapalene, showed that there was a statistically greater percentage reduction in TLs among participants on BPO/clindamycin plus adapalene than on adapalene (71% versus 52%; P < 0.05) at week 12. A statistically greater reduction was also found in NILs (71% versus 51%; P < 0.05). Despite no significant difference, BPO/clindamycin plus adapalene led to a greater reduction in ILs (71% versus 58%).

For the comparison between BPO/clindamycin plus tazarotene versus placebo plus tazarotene (Tanghetti 2006), the number of papules and pustules was reduced by 63% among participants on BPO/clindamycin and by 58% among participants on placebo at week 12, but no significant difference was found. The percentage reduction in open and closed comedones was 70% and 60% (P < 0.01).

Bowman 2005 reported that at week 10, the percentage reduction in TLs, ILs, and NILs in the BPO/clindamycin plus tretinoin and clindamycin treatment groups was 62.6%, 69.3%, and 61.0%, respectively, compared to 49.6%, 52.5%, and 46.2% in the other group. A significant difference was found in the percentage reduction in ILs (P = 0.02). For medium‐term and short‐term treatment duration, the percentage reduction in TLs (medium‐term: 55.8% versus 40.9%; short‐term: 49.8% versus 35.3%), ILs (53.7% versus 44.9%; 52% versus 45.7%), and NILs (59.6% versus 37.9%; 48.6% versus 21.9%) was consistently greater in the BPO/clindamycin add‐on treatment group, but it is unclear whether significant differences existed, as no information about this was reported.

Main pooled analyses: BPO/clindamycin versus placebo or no treatment

This outcome for the long term was reported in five trials (Bowman 2005; Eichenfield 2011; NCT01138514; Pariser 2014; Thiboutot 2008), in which 2226 participants were included in the BPO/clindamycin group and 1399 in the placebo group. The pooled risk ratio showed clear differences in the outcome (RR 2.37, 95% CI 1.87 to 3.01; 3993 participants; 5 studies; I² = 65%; Analysis 14.8), with substantial heterogeneity probably induced by the Bowman 2005 and NCT01138514 trials, in which the concentration of BPO was higher (5%) than in the others (2.5% or 3%).

Subgroup analyses: co‐intervention

We did not find that the effects of BPO on this outcome can differ by co‐intervention, with a P value of 0.25 (Analysis 14.8).

Studies not included in the meta‐analysis

Pariser 2014 reported this outcome as assessed by the acne‐specific quality of life (Acne‐QoL) questionnaire. There was no significant difference in the mean score for each domain at week 12, although each domain was scored higher in the BPO/clindamycin group (7.6 versus 5.9 in self‐perception, 5.5 versus 4.5 in role‐emotional, 4.7 versus 3.7 in role‐social, and 6.5 versus 4.3 in acne symptoms).

Main pooled analyses: BPO/clindamycin versus placebo or no treatment

With sufficient data for the meta‐analysis in eight trials, the pooled risk ratio was 0.91 (95% CI 0.78 to 1.07; 5042 participants; I² = 24%; Analysis 14.9). The most common adverse events observed across included trials were skin dryness, irritation, desquamation, and erythema.

Studies not included in the meta‐analysis

We cannot calculate RR as the effect size because the number of participants with at least one adverse event was not provided in the Del 2007 and Tanghetti 2006 trials. In Del 2007, the proportion of participants experiencing erythema, dryness, burning, and pruritus was 14.8% versus 24.3%, 51.6% versus 56.8%, 11.2% versus 12.6%, and 8.5% versus 15.5%. A significant difference was found in erythema (P < 0.05). Tanghetti 2006, the other trial, found no significant difference in the incidence of peeling and dryness between groups (10% versus 18% and 8% versus 12%, respectively). Trial authors reported no significant differences in grading for erythema, burning, dryness, or pruritus.

Main pooled analyses: BPO/clindamycin versus adapalene

Guerra‐Tapia 2012 included 83 participants in the BPO/clindamycin group and 85 participants in the adapalene group. The outcome was assessed with a 12‐grade system. The risk ratio of self‐rated improvement was 1.12 (95% CI 0.96 to 1.31) (83% versus 74%) at week 12 (Analysis 15.1).

Studies not included in the meta‐analysis

We cannot pool this result with those from the other trial ‐ Langner 2008 ‐ because the outcome was assessed on a different definition. In Langner 2008, 65 participants randomised to the BPO/clindamycin group and 65 to the adapalene group rated their condition as improved, no change, or worse. Trial authors reported that the percentage of improvement increased over time to about 90% at week 12 for both treatment groups. Although the proportion of improvement was greater early in treatment with BPO/clindamycin (weeks 1, 2, 4, and 8), no significant difference was found at week 12.

Main pooled analyses: BPO/clindamycin versus adapalene

The outcome was reported in three trials (Dubey 2016; Guerra‐Tapia 2012; Langner 2008). No evidence suggests that participants treated with BPO/clindamycin had higher risk of withdrawal due to adverse effects. The risk ratio was 0.41 (95% CI 0.05 to 3.05; 398 participants; 3 studies; I² = 0%; Analysis 15.2). Guerra‐Tapia 2012 did not specify any adverse events leading to treatment discontinuation. In Langner 2008, one participant in each group experienced adverse events leading to withdrawal from the study (BPO/clindamycin: erythema, burning, and pruritus; adapalene: dermatitis). No participants discontinued treatment due to adverse events in Dubey 2016.

Studies not included in the meta‐analysis

In Dubey 2016, trial authors found that mean NIL, IL, and total counts were reduced from 24.74 at baseline to 5.23 at week 12, from 16.38 to 2.12 and from 27.53 to 5.59, respectively, in the BPO/clindamycin group. Counterparts in the clindamycin group were from 25.91 to 3.95, from 16.25 to 2.00, and from 28.68 to 4.29, respectively. Trial authors stated that no statistical difference was found between two groups for any type of lesion count. Guerra‐Tapia 2012 found a significant difference in the reduction in TLs (‐33.9 versus ‐27.1; P < 0.05), ILs (‐17.0 versus ‐15.1; P < 0.05), and NILs (‐20.1 versus ‐18.0; non‐significance) at week 12. In another trial ‐ Ko 2009 ‐ investigators claimed that TLs (‐44.9 versus ‐34.1; P = 0.017) and ILs (‐29.7 versus ‐19.2; P = 0.026) were significantly reduced in 31 participants in the BPO/clindamycin group at week 12 compared to 38 participants in the adapalene group; however, this significant difference was not found in NILs (‐14.8 versus ‐15.0). Langner 2008 found that percentage reduction at week 12 was 70.4% versus 53.1% in TLs (P < 0.005), 81.5% versus 58.2% in ILs (P < 0.001), and 61.9% versus 50.0% in NILs (P < 0.005).

Main pooled analyses: BPO/clindamycin versus adapalene

Guerra‐Tapia 2012 reported this outcome as assessed with Skindex‐29. A significant difference was found for the long‐term (MD ‐4.20, 95% CI ‐7.06 to ‐1.34; 168 participants; Analysis 15.3) and short‐term periods (MD ‐3.80, 95% CI ‐6.16 to ‐1.44; 168 participants; Analysis 15.4) in favour of BPO/clindamycin.

Studies not included in the meta‐analysis

Twenty samples were collected from each group (BPO/clindamycin and adapalene groups) (Langner 2008). The reduction in total, erythromycin‐resistant, and clindamycin‐resistant C acnes strains at week 12 was ‐1.24 versus 0.50 versus ‐0.91 versus 0.13 and ‐0.80 versus 0.09 log₁₀ CFU cm^‐2. It is unclear whether the difference was significant.

Main pooled analyses: BPO/clindamycin versus adapalene

With a total of 148 participants in the BPO/clindamycin group and 150 in the adapalene group, the pooled risk ratio was 0.60 (95% CI 0.36 to 1.01; 298 participants; 2 studies; I² = 0%; Analysis 15.5), and the difference between groups was not statistically significant. In Guerra‐Tapia 2012, among 83 participants treated with BPO/clindamycin, site reactions, nasopharyngitis, and nervous system disorders occurred in five, eight, and five participants, respectively. In the adapalene group of 85 participants, counterparts were 14, 27, and 8, respectively. In Langner 2008, only one (1.5%) participant in the BPO/clindamycin group was considered to have treatment‐related adverse events (burning, erythema, and dryness). In the adapalene group, seven (10.8%) participants had adverse events (application site burning, dermatitis, desquamation, dryness, erythema, pain, and pruritus).

Studies not included in the meta‐analysis

In Dubey 2016, dryness was the only adverse event reported, which was observed in 2.1% of participants treated with adapalene, but dryness (2.2%) and itching (2.2%) were found in the BPO/clindamycin group. Ko 2009 reported that both drugs were well tolerated, with most mild adverse events observed within one month of treatments, such as erythema, desquamation, stinging/burning sensation, dry skin, and pruritus.

Main pooled analyses: BPO/clindamycin versus azelaic acid

Schaller 2016 included 108 participants in the BPO/clindamycin group and 109 in the azelaic acid group. The outcome was assessed via a 7‐point scale. Participants treated with BPO/clindamycin were more likely to achieve treatment success (rated as "very much improved" or "much improved") throughout the trial, with an RR of 1.38 (95% CI 1.05 to 1.81; Analysis 16.1), 1.44 (95% CI 1.09 to 1.89; Analysis 16.2), and 1.37 (95% CI 1.00 to 1.88; Analysis 16.3) for long‐, medium‐, and short‐term periods, respectively.

Main pooled analyses: BPO/clindamycin versus azelaic acid

One participant in each group discontinued treatment due to adverse effects (application site erythema, dryness, pruritus and pain, and left and right eyelid oedema occurred in the participant on BPO/clindamycin, and application site pruritus and pain occurred in the participant on azelaic acid) (Schaller 2016). No significant difference was observed between the two groups, with an RR of 1.01 (95% CI 0.06 to 15.93; 217 participants; Analysis 16.4).

Main pooled analyses: BPO/clindamycin versus azelaic acid

For the long‐term period, the mean difference was 18.50% (95% CI 10.54 to 26.46%; 211 participants; Analysis 16.5), 17.30% (95% CI 9.87 to 24.73%; 211 participants; Analysis 16.6), and 18.50% (95% CI 8.67 to 28.33%; 211 participants; Analysis 16.7) for TLs, ILs, and NILs, respectively.

For the medium‐term period, the counterpart was 15.10% (95% CI 7.19 to 23.01%; 206 participants; Analysis 16.8), 15.90% (95% CI 8.06 to 23.74%; 206 participants; Analysis 16.9), and 13.00% (95% CI 3.24 to 22.76%; 206 participants; Analysis 16.10).

The short‐term mean difference was 13.00% (95% CI 6.77 to 19.23%; 212 participants; Analysis 16.11), 14.10% (95% CI 6.18 to 22.02%; 212 participants; Analysis 16.12), and 11.10% (95% CI 3.56 to 18.64%; 212 participants; Analysis 16.13).

Main pooled analyses: BPO/clindamycin versus azelaic acid

This outcome was significantly different between the two groups for long‐ and medium‐term periods, with an RR of 1.91 (95% CI 1.17 to 3.11; Analysis 16.14) and 1.88 (95% CI 1.07 to 3.32; Analysis 16.15), respectively, in favour of BPO/clindamycin treatment. We cannot observe significant differences for the short‐term period (RR 1.74, 95% CI 0.87 to 3.49; Analysis 16.16).

Studies not included in the meta‐analysis

A greater percentage change in quality of life was reported at week 12 in the Children's Dermatology Life Quality Index (CDLQI) score (60.5% versus 36.8%) and the Dermatology Life Quality Index (DLQI) score (73.4% versus 31.9%) for BPO/clindamycin versus azelaic acid.

Main pooled analyses: BPO/clindamycin versus azelaic acid

Treatment‐related adverse events were reported in 13.9% and 33.0% of participants with BPO/clindamycin and azelaic acid treatment, respectively, with significantly lower risk in the BPO/clindamycin group (RR 0.42, 95% CI 0.24 to 0.72; 217 participants; Analysis 16.17). Local reactions occurred in 15.7% of participants treated with BPO/clindamycin (24 events in 17 participants) and 35.8% of participants with azelaic acid (60 events in 39 participants), including pruritus (7.4% versus 22.9%), pain (6.5% versus 20.2%), erythema (2.8% versus 4.6%), and dryness (1.9% versus 2.8%).

Studies not included in the meta‐analysis

Participants rated their improvement as improved, no change, or worse. There was no significant difference in the proportion of "improved" between the two groups (91.9% versus 89.3%) at week 12.

Main pooled analyses: BPO/clindamycin versus erythromycin/zinc

Two participants withdrew, with no significant differences between the two groups (RR 1.03, 95% CI 0.07 to 16.12; 148 participants; Analysis 17.1): one participant on BPO/clindamycin treatment due to rash and skin tightness and one on erythromycin/zinc treatment due to exacerbation of acne.

Main pooled analyses: BPO/clindamycin versus erythromycin/zinc

For the long‐term period, the mean difference was 6.10 (95% CI ‐4.65 to 16.85; Analysis 17.2), 0.10 (95% CI ‐3.85 to 4.05; Analysis 17.3), and 6.40 (95% CI ‐2.16 to 14.96; Analysis 17.4) for TLs, ILs, and NILs, respectively.

For the medium‐term period, the counterpart was 5.60 (95% CI ‐5.61 to 16.81; Analysis 17.5), ‐0.20 (95% CI ‐4.33 to 3.93; Analysis 17.6), and 6.20 (95% CI ‐2.66 to 15.06; Analysis 17.7).

The short‐term mean difference was 5.80 (95% CI ‐5.96 to 17.56; Analysis 17.8), 0.10 (95% CI ‐3.91 to 4.11; Analysis 17.9), and 6.20 (95% CI ‐3.60 to 16.00; Analysis 17.10).

Studies not included in the meta‐analysis

Only 14 participants treated with BPO/clindamycin and 15 treated with erythromycin/zinc contributed to the bacteriology data. At week 12, the total number of C acnes strains (log₁₀ CFU cm^‐2) was reduced from 5.38 (standard deviation (SD), 0.82) to 3.67 (SD, 1.30) and from 4.35 (SD, 1.19) to 3.49 (SD, 2.18) for BPO/clindamycin and erythromycin/zinc groups, respectively. In the BPO/clindamycin group, the number of erythromycin‐ and clindamycin‐resistant C acnes strains was reduced from 2.71 (SD, 2.35) to 1.45 (SD, 1.86) and from 2.48 (SD, 2.30) to 1.14 (SD, 1.75), respectively. In the erythromycin/zinc group, however, there was an increase in the number of erythromycin‐ and clindamycin‐resistant C acnes strains: from 0.76 (1.64) to 2.53 (SD, 2.05) and from 0.78 (SD, 1.56) to 2.19 (SD, 2.21), respectively. With no significance test results reported, trial authors claimed that the level of resistance appeared to decline with BPO/clindamycin but appeared to increase with erythromycin/zinc. We cannot estimate MD as the effect size because data were not sufficient to calculate the standard deviation of the change in C acnes strains from baseline.

Main pooled analyses: BPO/clindamycin versus erythromycin/zinc

A total of 20.5% of participants on BPO/clindamycin and 30.7% of participants on erythromycin/zinc reported at least one adverse event, with no significant difference detected (RR 0.67, 95% CI 0.38 to 1.18; 148 participants; Analysis 17.11). Common adverse events that occurred in each group were similar, including dryness (4.1% versus 5.3%), desquamation (4.1% versus 2.7%), burning (2.7% versus 2.7%), and erythema (2.7 versus 2.7%). There were no differences between treatments.

Studies not included in the meta‐analysis

The median percentage reduction was 65.0% versus 62.7% in TLs, 75.9% versus 70.7% in ILs, and 60.7% versus 55.3% in NILs at week 12. No significance test results were presented.

Main pooled analyses: BPO/clindamycin plus adapalene versus dapsone plus adapalene

No serious adverse events were found in both groups. During the 12‐week period, adverse events occurred in 15 participants on BPO/clindamycin plus adapalene (7 participants with nasopharyngitis and 8 with headache) and 23 participants on dapsone plus adapalene (15 participants with nasopharyngitis and 8 with headache), with no significant differences between the two groups (RR 0.67, 95% CI 0.37 to 1.23; Analysis 18.1).

Main pooled analyses: BPO/erythromycin versus placebo or no treatment

Jones 2002 presented data in accordance with the definition of treatment success in our review. The trial included 112 participants in the BPO/erythromycin group and 111 in the placebo group. The outcome was assessed with a 4‐point Likert‐like scale. The risk ratio of treatment success was 1.28 (95% CI 1.04 to 1.57; Analysis 19.1), supporting the benefits of BPO/erythromycin treatment.

Studies not included in the meta‐analysis

A similar scale was used in the other trial (Thiboutot 2002), but the proportion of treatment success defined in our review cannot be calculated based on data provided in the original report of this trial. With 245 participants treated with BPO/erythromycin and 82 with placebo, the trial found that participants receiving BPO/erythromycin had significantly greater scores at the end of treatment compared with participants receiving placebo (P < 0.001).

Another 12‐week parallel trial included 90 participants on BPO/erythromycin plus tazarotene and 89 participants on tazarotene alone (Draelos 2002). This outcome was assessed using a 5‐point Likert‐like scale (from highly favourable to highly unfavourable). At week 12, no significant difference was found in the proportion of participants rating the improvement as "highly favourable" or "favourable" (86% versus 73%).

Main pooled analyses: BPO/erythromycin versus placebo or no treatment

The long‐term trial reported that 6% of participants on BPO/erythromycin versus 11% on placebo discontinued treatments because of adverse events during the 12‐week period (Draelos 2002), with no clear differences detected (RR 0.49, 95% CI 0.18 to 1.39; 179 participants; Analysis 19.2). The trial did not specify which adverse events led to withdrawal.

For the other two medium‐term trials (Jones 2002; Thiboutot 2002), only one of 193 participants receiving placebo withdrew because of dryness and itching, and none withdrew from the BPO/erythromycin group consisting of 357 participants, with an RR of 0.33 (95% CI 0.01 to 8.02; 550 participants; 2 studies; I² = 0%; Analysis 19.3).

Studies not included in the meta‐analysis

Two trials provided long‐term data for this outcome (Chalker 1983; Sklar 1996). They included an equal number of participants in each group (76 participants in each) and found that no participants discontinued treatments because of adverse effects.

Main pooled analyses: BPO/erythromycin versus placebo or no treatment

For the long‐term change in lesion counts, the mean difference in the absolute change was ‐5.20 (95% CI ‐11.74 to 1.34; Analysis 19.4), ‐6.93 (95% CI ‐11.55 to ‐2.31; Analysis 19.5), and 1.80 (95% CI ‐1.96 to 5.56; Analysis 19.6) for TLs, ILs, and NILs, respectively (Sklar 1996).

For the medium‐term change in lesion counts, the mean difference in the absolute change pooled from two trials was ‐4.59 (95% CI ‐12.63 to 3.44; 281 participants; 2 studies; I² = 75%; Analysis 19.7), ‐3.90 (95% CI ‐8.07 to 0.27; 58 participants; 1 study; Analysis 19.8), and 0.80 (95% CI ‐4.30 to 5.89; 281 participants; 2 studies; I² = 73%; Analysis 19.9) for TLs, ILs, and NILs respectively, with no significant differences but high heterogeneity (Jones 2002; Sklar 1996).

Sklar 1996 also reported the short‐term outcome. It found no significant differences between the two groups in the absolute change in TLs (MD ‐3.60, 95% CI ‐7.92 to 0.72; Analysis 19.10) and in NILs (MD 2.40, 95% CI ‐0.71 to 5.51; Analysis 19.12). However, the change in ILs differed between groups, favouring the BPO/erythromycin group (MD ‐6.10, 95% CI ‐9.39 to ‐2.81; Analysis 19.11).

Studies not included in the meta‐analysis

The other long‐term trial mentioned only whether the significant difference was found (Chalker 1983). As reported, BPO/erythromycin was significantly superior to placebo at week 10 in the percentage reduction in comedone, pustule, papule, and inflammatory lesions.

We cannot obtain mean difference from another medium‐term trial (Thiboutot 2002). Trial authors claimed that significant differences in the percentage reduction in TLs, ILs, and NILs were found at week 8. In the BPO/erythromycin group, TLs were reduced by 58.5%, ILs by 62.6%, and NILs by 59%, compared to 27.5%, 34.1%, and 28.0% in the placebo group.

For the comparison between BPO/erythromycin plus tazarotene and tazarotene, the percentage reduction in ILs was significantly different (65.2% versus 43.7%; P < 0.05) between two groups at week 12 but not the reduction in NILs (48.4% versus 41.6%; P > 0.05). For the medium‐ and short‐term periods, both treatments reduced NILs by about 40% and 30%, respectively, with no significant differences. On the other hand, BPO/erythromycin plus tazarotene was more effective for reducing ILs (58% versus 31% for the medium‐term period and 44% versus 17% for the short‐term period). We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Main pooled analyses: BPO/erythromycin versus placebo or no treatment

This outcome was reported in two medium‐term trials (Jones 2002; Thiboutot 2002), in which 357 and 193 participants were randomised to BPO/adapalene and placebo groups, respectively. The pooled risk ratio was 2.84 (95% CI 1.79 to 4.52; I² = 9%; Analysis 19.13). For the short‐term outcome, Thiboutot 2002 reported the risk ratio of 6.69 (95% CI 0.91 to 49.10; Analysis 19.14), and Jones 2002 mentioned significant differences (P < 0.021).

Main pooled analyses: BPO/erythromycin versus placebo or no treatment

For the long‐term outcome, with no significant differences between two groups, 15 of 90 versus 14 of 89 participants had at least one treatment‐related adverse event (RR 1.06, 95% CI 0.54 to 2.06; 179 participants; Analysis 19.15). The distribution of adverse events in each group is unclear. Chalker 1983 reported no adverse events observed during a 10‐week period. The other long‐term trial did not present the total number of participants with at least one adverse event but claimed that the incidence of scaling and tightness in the BPO/erythromycin group was significantly higher (P < 0.05) than in the placebo group.

In the two medium‐term trials (Jones 2002; Thiboutot 2002), the risk ratio was 1.88 (95% CI 0.92 to 3.87; Analysis 19.16).

Studies not included in the meta‐analysis

This outcome was assessed in Draelos 2002 via a 5‐point Likert‐like scale (from highly favourable to highly unfavourable). At week 12, no significant difference was found in the proportion of participants rating improvement as highly favourable or favourable (86% versus 85%). We cannot estimate RR as the effect size because data were not sufficient to calculate the number of participants with treatment success defined in our review.

Main pooled analyses: BPO/erythromycin versus clindamycin

Participants on BPO/erythromycin tended to discontinue treatments because of adverse events, with no significant differences between the two groups (RR 1.51, 95% CI 0.50 to 4.61; 376 participants; 2 studies; I² = 0%; Analysis 20.1). Both trials did not report which adverse events led to their withdrawal.

Subgroup analyses: co‐intervention

No evidence suggests the co‐intervention may affect the effect, with a P value of 0.49 (Analysis 20.1).

Studies not included in the meta‐analysis

We cannot estimate MD as the effect size because data in both trials were not sufficient to calculate standard deviation.

In Packman 1996, the absolute change in ILs and NILs was significantly different between the two groups at any time point, in favour of BPO/erythromycin. The mean number of ILs decreased from 18.2 at baseline to 11.5 (short term), 9.0 (medium term), and 8.4 (long term) in participants on BPO/erythromycin, and the counterpart decreased from 17.4 to 12.7, 12.0, and 12.3 in participants on clindamycin. The mean number of NILs decreased from 25.6 at baseline to 15.4, 14.7, and 13.6 in participants on BPO/erythromycin, but there was almost no change in participants on clindamycin (from 23.6 to 25.5, 22.6, and 23.2).

In Draelos 2002, the percentage reduction in ILs at week 12 was significantly different (65.2% versus 40.1%; P < 0.05) but not in NILs (48.4% versus 55.7%; P > 0.05). For the medium‐ and short‐term periods, both treatments reduced NILs by about 40% and 30%, respectively, with no significant differences. On the other hand, BPO/erythromycin plus tazarotene seemed more effective for reducing ILs (58% versus 35% for the medium‐term period and 44% versus 22% for the short‐term period).

Main pooled analyses: BPO/erythromycin versus clindamycin

Non‐significantly increased risk of adverse event occurrence was observed between the two groups (RR 2.16, 95% CI 0.77 to 6.08; 378 participants; 2 studies; I² = 23%; Analysis 20.2). Adverse events reported in the trials included dryness, erythema, pruritus, stinging, burning, and unusual taste and smell.

Subgroup analyses: co‐intervention

No evidence suggests that the co‐intervention may affect the effect, with a P value of 0.25 (Analysis 20.2).

Studies not included in the meta‐analysis

Trial authors reported that BPO/erythromycin was superior to azelaic acid in reducing ILs at week 8, but they were not significantly different in reducing NILs. We cannot estimate RR as the effect size because insufficient data were reported in the abstract.

Studies not included in the meta‐analysis

At week 4, the difference in TLs between the two treatment groups was not significant. The mean number of TLs decreased from of 18.1 at baseline to 3.7 in participants on BPO/erythromycin and from 17.8 to 4.2 in participants on metronidazole. We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Studies not included in the meta‐analysis

No participants discontinued treatments due to adverse effects.

Main pooled analyses: BPO/erythromycin versus viaminate

No adverse events occurred in the viaminate group, but 13 participants in the BPO/erythromycin group had dryness and mild peeling and redness, with significantly higher risk in the BPO/erythromycin group (RR 23.52, 95% CI 1.42 to 390.03; 187 participants; Analysis 21.1).

Studies not included in the meta‐analysis

Participants rated overall acne improvement using a 6‐point scale (0 = worse, 1 = no change, 2 = slight improvement, 3 = mild improvement, 4 = moderate improvement, 5 = excellent improvement). For this outcome, trial authors reported only that participant self‐assessment was strongly in favour of BPO/erythromycin (P < 0.001).

Studies not included in the meta‐analysis

No participants on BPO/erythromycin and three participants on zinc/erythromycin discontinued treatment due to adverse events. It is unclear what adverse events occurred in these three participants.

Studies not included in the meta‐analysis

A significant difference between groups was found in the percentage reduction in both ILs (86.6% versus 60.3%) and NILs (53.1% versus 26.8%) for the long‐term period (week 10) (both P values < 0.005). For the medium‐term period, significant differences were observed in both ILs (80% versus 52%) and NILs (49% versus 25%). However, the percentage reduction in ILs was significantly different for the short‐term period (69% versus 44%), but not the percentage reduction in NILs (26% versus 14%).

Studies not included in the meta‐analysis

Only one participant on BPO/erythromycin reported adverse events (dryness and irritation). In the zinc/erythromycin group, four participants reported nine adverse events, including irritation, itching, dryness, redness, scratching, scaling, and soreness.

Studies not included in the meta‐analysis

Researchers assessed the outcome using a 4‐point Likert‐like scale (from worse = ‐1 to greatly improved = 2). The average score was 1.15 and 0.53 for the BPO/sulphur and placebo groups, respectively. We cannot estimate RR as the effect size because data were not sufficient to calculate the number of participants with treatment success as defined in our review.

Studies not included in the meta‐analysis

The change in lesion counts was assessed for superficial (comedones and pustules) and deep (papules and cysts) lesions, respectively, via an 11‐point scoring system (decrease in lesions by 100% = 5, 76% to 99% = 4, 51% to 75% = 3, 26% to 50% = 2, < 25% = 1, and no change = 0; increase by < 25% = ‐1, 26% to 50% = ‐2, 51% to 75% = ‐3, 76% to 100% = ‐4, and over 100% = ‐5). The average score for superficial lesions was 0.81 and 0 for the BPO/sulphur and placebo groups, respectively, and the counterpart for deep lesions was 0.91 and 0.53. We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Studies not included in the meta‐analysis

Individual adverse events were reported, but the total number of participants with any adverse event was not reported. Excessive erythema and dryness occurred in two participants on placebo and in four participants taking BPO/sulphur treatment.

Studies not included in the meta‐analysis

No participants discontinued treatment because of any adverse events.

Main pooled analyses: BPO/glycolic acid/zinc lactate versus placebo

For the long‐term change in lesion counts, the absolute change in TLs and ILs was significantly different, with an MD of ‐6.50 (95% CI ‐12.56 to ‐0.44; 56 participants; Analysis 22.1) and ‐5.60 (95% CI ‐10.38 to ‐0.82; 56 participants; Analysis 22.2), respectively. No clear difference in NILs was found (MD ‐0.90, 95% CI ‐4.74 to 2.94; 56 participants; Analysis 22.3).

For the medium‐term absolute change in lesion counts, no significant difference was observed for any types of lesions, with an MD of ‐0.90 (95% CI ‐5.26 to 3.46; 56 participants; Analysis 22.4), ‐2.60 (95% CI ‐6.59 to 1.39; 56 participants; Analysis 22.5), and 1.70 (95% CI ‐1.95 to 5.35; 56 participants; Analysis 22.6) for TLs, ILs, and NILs, respectively.

Trial authors also reported the short‐term outcome, noting no significant differences between the two groups in the absolute change in TLs (MD ‐4.20, 95% CI ‐9.29 to 0.89; 58 participants; Analysis 22.7) nor in NILs (MD 0.50, 95% CI ‐3.41 to 4.41; 58 participants; Analysis 22.9). However, the change in ILs differed, favouring the BPO/erythromycin group (MD ‐4.80, 95% CI ‐8.66 to ‐0.94; 58 participants; Analysis 22.8).

Studies not included in the meta‐analysis

Trial authors reported that adverse events including mild erythema, scaling, tightness, burning, stinging, itching, or tingling were observed in both groups. However, the number of participants with any event in each group was not reported.

Main pooled analyses: BPO/potassium hydroxyquinoline sulphate versus placebo

Three participants on BPO/potassium hydroxyquinoline sulphate discontinued treatment because of stinging or skin irritation, with no significant differences between the two groups (RR 7.81, 95% CI 0.42 to 144.12; 53 participants; Analysis 23.1).

Studies not included in the meta‐analysis

Trial authors mentioned that adverse events were recorded; however, no information on specific adverse events was available in the results section, except that three participants withdrew due to adverse events.

Main pooled analyses: BPO (10%) versus BPO (5%)

One of 93 participants treated with 10% BPO discontinued treatment due to adverse effects, and five of 164 participants in the 5% BPO group withdrew (Ji 2000; Wang 2003). Such differences were not significant, with an RR of 0.40 (95% CI 0.06 to 2.52; I² = 0%; Analysis 24.1). It is unclear which adverse events resulted in their withdrawal.

Studies not included in the meta‐analysis

Neither of the trials reported the change in lesion counts from baseline, but trial authors reported lesion counts at the end of the study. Compared with 5% BPO, the mean difference in IL counts at week 6 was ‐0.62 (95% CI ‐1.97 to 0.73; I² = 19%), and the counterpart in NIL counts was ‐7.56 (95% CI ‐12.04 to ‐3.07; I² = 65%).

Main pooled analyses: BPO (10%) versus BPO (5%)

No evidence suggests a significant difference in the occurrence of adverse events. The risk ratio was 0.72 (95% CI 0.40 to 1.31; I² = 65%; Analysis 24.2).

Studies not included in the meta‐analysis

In total, the two trials included 58 participants treated with 10% BPO and 56 treated with 2.5% BPO. None in both groups discontinued treatment due to adverse effects.

Studies not included in the meta‐analysis

With 24 participants treated with 10% BPO and an equal number of participants with 2.5% BPO, Mills 1986 found similar percentage reduction in ILs (44.7% versus 46.7%) at week 8.

In Wang 2003, 33 participants were included in the 10% BPO group and 124 in the 2.5% BPO group. Trial authors presented lesion counts at week 6, rather than the change in lesion counts from baseline, suggesting that both IL and NIL counts were fewer in the 10% BPO group (MD for ILs ‐3.80, 95% CI ‐5.99 to ‐1.61; MD for NILs ‐3.70, 95% CI ‐6.66 to ‐0.74).

Main pooled analyses: BPO (10%) versus BPO (2.5%)

In Wang 2003, more participants in the 10% BPO group were experiencing adverse events (20/33 versus 11/31; RR 1.71, 95% CI 0.99 to 2.96; Analysis 25.1). Mills 1986 reported a significant difference in the frequency and severity of erythema, peeling, and burning between those receiving 10% BPO and those receiving 2.5% BPO at all follow‐up visits.

Studies not included in the meta‐analysis

The outcome was assessed in Dhawan 2013 (including 20 participants in each group) via a 5‐point scale (1 = very satisfied, 2 = satisfied, 3 = neutral, 4 = unsatisfied, 5 = very unsatisfied). There was no significant difference between the two groups in the proportion of participants rating the improvement as "very satisfied" or "satisfied" (88.2% versus 85.0%) at week 12. We cannot estimate RR as the effect size because data were not sufficient to calculate the number of participants with treatment success defined in our review.

Main pooled analyses: BPO (5%) versus BPO (2.5%)

In the long‐term trials (Dhawan 2013; Kawashima 2017a; Kawashima 2017b), the pooled RR of this outcome was 1.28 (95% CI 0.65 to 2.54; 906 participants; I² = 0%; Analysis 26.1). Adverse events leading to withdrawal included application site erythema and irritation.

Wang 2003, a six‐week trial, and Mills 1986, an eight‐week trial, reported medium‐term outcomes. Among a total of 151 participants treated with 5% BPO and 57 participants treated with 2.5% BPO, only three participants on 5% BPO treatment discontinued treatment due to adverse effects (Mills 1986; Wang 2003), with an RR of 1.79 (95% CI 0.09 to 33.82; Analysis 26.2). Trial authors did not specify which adverse events led to their withdrawal.

Subgroup analyses: co‐intervention

For the long‐term outcome, no evidence suggests that the effect may be affected by the co‐intervention (P = 0.59; Analysis 26.1).

Studies not included in the meta‐analysis

In Yong 1979, in which participants were treated for 4 to 18 weeks, 11 participants did not complete the trial because of desquamation and erythema, and four participants with pustulation and exacerbation of acne lesions were withdrawn. In this trial, however, it is unclear how many participants in each group withdrew.

Studies not included in the meta‐analysis

With 25 participants on 5% BPO and 26 on 2.5% BPO, Mills 1986 found similar percentage reduction in ILs (57.7% versus 55.9%) at week 8. A total of 124 participants were included in the 5% BPO group and 31 in the 2.5% BPO group (Wang 2003). Trial authors presented lesion counts at week 6, rather than the change in lesion counts from baseline, suggesting that IL but not NIL counts were lower in the 5% BPO group (MD for ILs ‐4.00, 95% CI ‐6.42 to ‐1.58; MD for NILs 1.80, 95% CI ‐1.69 to 5.29). In a 12‐week trial including 78 participants on 5% BPO and 79 on 2.5% BPO (NCT02073461), the reduction in TLs was 65.9% and 58.3%, respectively, with no significance test results presented. In Yong 1979, 98 participants on 5% BPO and 96 on 2.5% BPO were assessed for treatment improvement based on the percentage reduction in lesion counts. At least 50% reduction was found in 74 versus 64 participants. Dhawan 2013 suggested no significant difference in the percentage reduction in TLs, ILs, and NILs at week 12 (67.7% versus 67.9%, 66.1% versus 53.8%, and 67.8% versus 75.0%, respectively). We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Main pooled analyses: BPO (5%) versus BPO (2.5%)

In Dhawan 2013, no significant difference was found in the mean change in Skindex‐29 global scores from baseline at week 12 (MD ‐2.40, 95% CI ‐8.68 to 3.88; 40 participants; Analysis 26.3).

Main pooled analyses: BPO (5%) versus BPO (2.5%)

Three studies provided long‐term data for this outcome in a total of 1063 participants (529 in the 5% group and 534 in the 2.5% group), with no clear differences between groups (RR 1.06, 95% CI 0.95 to 1.19; Analysis 26.4) (Dhawan 2013, Kawashima 2017a; Kawashima 2017b; NCT02073461).

Kawashima 2017a reported that the top three adverse events were skin exfoliation (52/227 versus 42/231), application site irritation (46/227 versus 44/231), and application site erythema (41/227 versus 32/231). Similarly, the most common adverse event in Kawashima 2017b was skin exfoliation (48/204 versus 39/204), followed by erythema (22/204 versus 28/204) and irritation (25/204 versus 17/204). Skin irritation (1/78 versus 3/79) was the most frequent skin or cutaneous adverse event reported in NCT02073461. Dhawan 2013 reported that 11 participants on BPO (5%)/clindamycin and seven participants on BPO (2.5%)/clindamycin had at least one adverse event, most of which were related to infections and infestations with nasopharyngitis.

Three trials reported the medium‐term outcome (Mills 1986; Wang 2003; Yong 1979); of these, only Wang 2003 provided sufficient data for an effect estimate. More participants in the 5% BPO group experienced adverse events (82/124 versus 11/31; RR 1.86, 95% CI 1.14 to 3.05; Analysis 26.5).

Subgroup analyses: co‐intervention

For the long‐term outcome, we did not find evidence suggesting this effect may differ by co‐intervention, with a P value of 0.86 (Analysis 26.4).

Studies not included in the meta‐analysis

Two medium‐term trials did not provide sufficient data for the meta‐analysis. Mills 1986 reported no significant differences in the frequency and severity of erythema, peeling, and burning between those receiving 5% BPO and those given 2.5% BPO. Adverse events were reported individually in Yong 1979, with erythema the most common event (50/98 versus 45/96), followed by desquamation (28/98 versus 22/96).

Studies not included in the meta‐analysis

The outcome was assessed via a 5‐point scale (from 0 = no improvement or worsening to 4 = complete clearing). With no significance test results reported, trial authors claimed that more participants on 5.5% BPO cream rated the improvement as "marked" or "better" than those on 6% BPO gel (38% versus 32%). We cannot estimate the RR because data were not sufficient to calculate the number of participants with treatment success defined in our review.

Main pooled analyses: BPO gel (6%) versus BPO cream (5.5%)

One participant on 6% BPO gel discontinued treatment due to application site irritation, with no significant differences between the two groups (RR 3.00, 95% CI 0.13 to 70.16; 48 participants; Analysis 27.1).

Studies not included in the meta‐analysis

No significant difference was found in the percentage reduction in TLs, ILs, and NILs at week 12 (21.2% versus 35.6%, 25.1% versus 39.3%, and 19.5% versus 33.0%, respectively). We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Studies not included in the meta‐analysis

As reported by trial authors, 46 adverse events were reported in 48% of participants and were more frequently observed in participants on 6% BPO. However, the number of adverse events for each group is unclear. The most common events were dryness, erythema, scaling, itching, stinging, and burning.

Studies not included in the meta‐analysis

Trial authors found no significant differences in the reduction in TLs and ILs between groups during the eight weeks. We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Studies not included in the meta‐analysis

Trial authors reported that no statistical differences in the percentage change in comedones (62% versus 57%), papules (56% versus 56%), pustules (80% versus 79%), and total lesions (62% versus 59%) were found between groups at week 8. We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Studies not included in the meta‐analysis

No adverse events were reported in either of the BPO gel groups.

Studies not included in the meta‐analysis

The trial registry published papule and pustule counts for each individual at any time point, with no statistical results available.

Studies not included in the meta‐analysis

No serious adverse events were found in the 10 participants. During the four‐week period, adverse events occurred in eight participants. All experienced burning/stinging.

Studies not included in the meta‐analysis

Six participants discontinued treatment due to local side effects. Two participants had severe dermatitis on both sides. Side effects occurred on both sides in another two participants but were worse on the side treated with alcohol‐based BPO: one participant had erythema, itching, and stinging, and the other had erythema and scaling. Another two participants stopped alcohol‐based BPO because of severe stinging.

Studies not included in the meta‐analysis

Trial authors reported no significant differences (mean difference was null) at each time point (four weeks and eight weeks) for any types of lesions including open comedones, closed comedones, pustules, papules, and cysts. We cannot estimate MD as the effect size because data were not sufficient to calculate standard deviation.

Studies not included in the meta‐analysis

No significant difference in the incidence of peeling and dryness was found between groups (10% versus 18% and 8% versus 12%, respectively). Trial authors reported no significant differences in grading for erythema, burning, dryness, or pruritus.